Prediction of oncogene mutation status in non-small cell lung cancer: A systematic review and meta-analysis with a special focus on artificial-intelligence-based methods

Almudena Fuster-Matanzo, Alfonso Picó Peris, Fuensanta Bellvís Bataller, Ana Jimenez-Pastor, Glen J. Weiss, Luis Martí-Bonmatí, Antonio Lázaro Sánchez, Giuseppe L. Banna, Alfredo Addeo, Ángel Alberich-Bayarri
doi: https://doi.org/10.1101/2024.05.31.24308261

ABSTRACT

Background In non-small cell lung cancer (NSCLC), alternative strategies to determine patient oncogene mutation status are essential to overcome some of the drawbacks associated with current methods. We aimed to review the use of radiomics alone or in combination with clinical data and to evaluate the performance of artificial intelligence (AI)-based models on the prediction of oncogene mutation status.

Methods A PRISMA-compliant literature review was conducted. The Medline (via Pubmed), Embase, and Cochrane Library databases were searched for studies published through June 30, 2023 predicting oncogene mutation status in patients with NSCLC using radiomics. Independent meta-analyses evaluating the performance of AI-based models developed with radiomics features or with a combination of radiomics features plus clinical data for the prediction of different oncogenic driver mutations were performed. A meta-regression to analyze the influence of methodological/clinical factors was also conducted.

Results Out of the 615 studies identified, 89 evaluating models for the prediction of epidermal growth factor-1 (EGFR), anaplastic lymphoma kinase (ALK), and Kirsten rat sarcoma virus (KRAS) mutations were included in the systematic review. A total of 38 met the inclusion criteria for the meta-analyses. The AI algorithms’ sensitivity/false positive rate (FPR) in predicting EGFR, ALK, and KRAS mutations using radiomics-based models was 0.753 (95% CI 0.721–0.783)/0.346 (95% CI 0.305–0.390), 0.754 (95% CI 0.639–0.841)/ 0.225 (95% CI 0.163–0.302), and 0.744 (95% CI 0.605–0.846)/0.376 (95% CI 0.274–0.491), respectively. A meta-analysis of combined models was only possible for EGFR mutation, revealing a sensitivity/FPR of 0.800 (95% CI 0.767–0.830)/0.335 (95% CI 0.279–0.396). No statistically significant results were obtained in the meta-regression.

Conclusions Radiomics-based models may represent valuable non-invasive tools for the determination of oncogene mutation status in NSCLC. Further investigation is required to analyze whether clinical data might boost their performance.

INTRODUCTION

Lung cancer represents the most often diagnosed cancer in both women and men worldwide, ranking first and third, respectively, and remaining the leading cause of cancer death1. Non-small cell lung cancer (NSCLC), the most frequent histological subtype, accounts for 80%–85% of cases, being adenocarcinoma the most common subtype (40%–50% of cases). Adenocarcinoma can be further subdivided into distinct molecular subtypes2. Indeed, molecular subtyping has become highly relevant in the disease context, as genotype-driven therapy (“targeted therapy”) is nowadays the standard of care for a significant subgroup of patients with advanced and metastatic NSCLC3. However, traditional methods for determining the molecular genotype, as well as the possible emergence of drug resistance mutations during patient’s follow-up, entail invasive biopsies and genetic sequence testing, procedures with multiple number of associated drawbacks including high costs, sampling bias, lack of enough sample, turnaround time, and medical complications4-6. Importantly, the overall accessibility of molecular diagnostics and liquid biopsy may be limited for many patients7, highlighting the need to investigate complementary methods to characterize the oncogene mutation status of lesions.

Radiological imaging represents a potent non-invasive tool for lung cancer, from the screening, diagnosis and staging of the disease to the management, therapeutic planification and follow-up of both early- and advanced-stage cases8. Specifically, computed tomography (CT) remains the standard of care for lung cancer visualization, providing excellent morphological and textural information. In recent years, radiomics, the process of extracting and analyzing quantitative features from medical images to investigate potential connections with biology and clinical outcomes, has gained increasing attention for its applicability in several oncological diseases including lung cancer8. The application of artificial intelligence (AI) to imaging analyses has enabled important clinical needs to be met. This includes the prognostication of outcomes or the prediction of response to treatment, disease progression, or the mutational and molecular profiling of tumors9. In particular, the use of radiomics coupled with AI methods has demonstrated to be a promising non-invasive alternative tool for the prediction of oncogene mutation status in NSCLC8.

In this systematic review and meta-analysis we aimed to: 1) review the available scientific evidence on the use of imaging-based models and radiomics for the prediction of the main targetable oncogenic driver alterations in NSCLC, including epidermal growth factor receptor (EGFR), anaplastic lymphoma kinase (ALK), and Kirsten rat sarcoma virus (KRAS); 2) analyze the overall performance of specifically, AI-based methods, for the prediction of oncogene mutation status; 3) evaluate whether the inclusion of clinical variables in the models improve their performance; 4) evaluate the impact of the available evidence from a clinical perspective.

MATERIAL AND METHODS

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines10. The review was registered on PROSPERO before initiation (registration no. CRD42022349809).

Search strategy

A systematic search for eligible publications published through 30 June 2023 was performed in Medline (via Pubmed), Cochrane Library and EMBASE databases using the keywords “Radiomics”, “NSCLC” and “Mutational status”. Further details on the search terms used in each database are provided in Supplementary Table S1. There were no limitations on the publishing year, participant age, or nationality. The search was exclusively limited to English-language publications.

Study selection

Literature search and study selection were independently performed by two reviewers (A.F.M. and A.L.S.). To find relevant publications, they reviewed the titles and abstracts. Studies that satisfied the inclusion criteria were then manually assessed for eligibility by full-text screening. Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia. Available at www.covidence.org) was used as a screening and data extraction tool.

Inclusion criteria

Papers were included in the qualitative synthesis (systematic review) if meeting the following inclusion criteria based on Patient, Index test, Comparator, Reference test, Diagnosis of reference (PIRD) questions: 1) being focused on the ability of radiomics to predict oncogene mutation status in NSCLC; 2) radiomics features were extracted from CT or from F-18 fluoro-deoxy-glucose (FDG)/CT scans; 3) a full text was available; 4) were written in English.

Exclusion criteria

Papers describing studies conducted using MRI scans (not the standard of care for NSCLC patients) or performed in phantom or animal models, or published as case reports, editorials, reviews, poster presentations, letters, editorials, or meeting abstracts were excluded. Papers not on the field of interest were also excluded.

For the quantitative synthesis (meta-analysis), the following additional exclusion criteria were applied: 1) oncogene mutation status was not the primary objective of the paper; 2) were focused on specific mutation subtypes; 3) did not apply AI-based methodologies; 4) developed simultaneous detection models or discriminant models; 5) sensitivity or specificity metrics were not available and could not be calculated; 6) were not comparable with the other articles included (model was developed based on intra- and extra-tumor derived radiomics features); 7) only included models developed with a combination of quantitative features extracted from PET/CT or from PET images (strictly adhering to a clinical perspective, PET scanning equipment is not always available and CT remains the standard of care for NSCLC patients); 8) did not reach a sufficient quality score according to the quality assessment (described below).

Quality assessment

The methodological quality of each study for its possible inclusion in the quantitative assessment was evaluated by using the Checklist for Artificial Intelligence in Medical Imaging (CLAIM)11. Classification, image reconstruction, text analysis, and workflow optimization are some of the applications of AI in medical imaging that are addressed by CLAIM, which is modeled after the Standards for Reporting of Diagnostic Accuracy Studies (STARD) guideline12-15. CLAIM checklist consists of 42 items divided into the conventional sections included in peer-reviewed scientific articles: title or abstract (1 item), abstract (1 item), introduction (2 items), methods (28 items subdivided into study design [2 items], data [7 items], ground truth [5 items], data partitions [3 items], model [3 items], training [3 items] and evaluation [5 items]), results (5 items subdivided into data [2 items] and model performance [3 items]), discussion (2 items) and other information (3 items). The CLAIM guideline offers a roadmap for writers and reviewers with the intention of fostering clear, open, and verifiable scientific discourse on the use of AI in medical imaging11.

For our quality assessment, a score was calculated for each paper ([total score, 42 - number of “not applicable” fields in each case]). A cut-off value of at least half of the total score after removing the “not applicable” items was established for the inclusion in the quantitative analysis. Therefore, this cut-off value varied for each study depending on the number of items that were applicable from among the 42 total items included in the CLAIM checklist (e.g., a cut-off value of 19 was established for those studies in which only 38 items of the checklist were applicable). See Supplementary Table S2. The assessment of the rigor, quality, and generalizability of the work of all enrolled studies was performed by three reviewers (A.J.P., F.B.B. and A.P.P.).

Data extraction

Data extracted included the following: (1) study details: first author, publication year, research questions, study design; (2) patient details: the source of data acquisition (single-center/multicenter), sample size, smoking history, age, sex, TNM staging, treatment status (naïve or any treatment received prior image acquisition), histological subtype; (3) imaging details: imaging modality, plain or contrast CT; (4) oncogene mutation status-related information: type of mutation, specific subtype of mutation (if available), sequencing method; sequencing kit (5) radiomics details: segmentation software, type of segmentation (manual, automatic, or semi-automatic), radiomics feature extraction software, number of imaging features extracted, number and name of radiomics features included in final models, features selection methods, type of models constructed (machine learning [ML], deep learning [DL], classical statistical model), final classifier used in machine learning models, clinical variables included in the models (if applicable), and models performance. Two independent reviewers (A.F.M. and A.L.S.) completed the initial screening and extracted data from all included studies.

Data analysis

For studies including models based on features extracted from different imaging modalities, only those based on CT scans were included in the quantitative analysis. A bivariate analysis of sensitivity and specificity as proposed by Reitsma et al.16 was chosen to perform the meta-analyses. This method has the distinct advantage of preserving the two-dimensional nature of the underlying data. It can also produce summary estimates of sensitivity and specificity (false positive rate [FPR, 1-specificity]), recognizing any possible correlation between these two measures. The method uses a random effect approach in which the values of the sensitivity and FPR estimates are obtained with restricted maximum likelihood. As a complement to the bivariate approach, the summary receiver operating characteristic (sROC) was calculated by converting each pair of sensitivity and specificity into a single measure of accuracy, the diagnostic odds ratio (DOR).

The analyses were carried out by reproducing the confusion matrices of each model presented in the studies, the number of cases and the prevalence of oncogene mutant positive cases. All calculations were performed on the basis of validation cohorts for studies applying a training/validation split method, or on the basis of the total sample when cross-validation was the validation strategy. To ensure homogeneity, calculations were conducted based on internal validation cohort data when external validation was also performed (minority of the cases).

Finally, a meta-regression analysis was performed to measure the possible influence of the following predictors: (1) average age of the cases, (2) manual segmentation vs semi-automatic segmentation vs both procedures (no studies including automatic segmentation approaches met the inclusion criteria for the quantitative analysis), (3) whether the model included only radiomics features or was combined with clinical variables, and (4) whether the model was classified as ML or DL. The heterogeneity in the description of the clinical variables included in the models prevented the inclusion of additional predictors of greatest clinical interest. Only the best model from each study according to its DOR was selected. When the mean/median age was not available due to the heterogeneity among studies when presenting descriptive results, it was inferred from the information obtained. Thus, mean and median values were indistinctly considered; when both values were provided, an average of both was calculated. If mean values were absent, median values were considered and viceversa. If both values were absent from the validation cohort, mean/median age from the total cohort was considered. When this information was not available either, the study was not included in the meta-regression.

All the analyses were performed using R Statistical Software v4.2.2 and the packages mada and tidyverse.

RESULTS

In total 615 articles were obtained according to the search strategy (Figure 1). After de-duplication, 397 studies were obtained and screened. According to the inclusion and exclusion criteria, 89 studies were included in the qualitative analysis (systematic review), all of them developing models for the prediction of EGFR, ALK, and/or KRAS. Out of those, 38 were found eligible for the quantitative part of the study (meta-analyses). As detailed in Supplementary Table S2, all papers passed this quality check and were therefore included.

Figure 1.
Figure 1.

PRISMA flowchart. AI, artificial intelligence; CLAIM, Checklist for Artificial Intelligence in Medical Imaging; CT, computed tomography; MRI, magnetic resonance imaging.

Qualitative analysis (systematic review)

Methodological characteristics of the studies

The methodological characteristics of the studies are summarized in Table 1. Most of the studies (n = 69/ 89) applied exclusively ML algorithms, while this methodology was also used to build comparator models in 10 articles in which DL techniques were the main methodological approach followed. Only three studies exclusively applied DL algorithms, while classical statistical models were used in seven publications. Among the 79 articles applying ML techniques, the most common classifier used was logistic regression (n = 38), followed by support vector machine (n = 35) and by random forest (n = 29). In terms of partitioning strategy, training-validation split was the most frequent technique (n = 71). External validation was only performed in a small set of studies (n = 9). Regarding imaging techniques, CT was the most frequently used modality (n = 61), followed by PET/CT (n = 22) and by PET alone (n = 4). Additionally, in one study17 PET/CT scans and contrast-enhanced CT images independently acquired were collected, while in another study18, PET/CT, CT, and contrast-enhanced diagnostic quality (CTD) images were used. Of the 61 studies conducted with CT scans, 39 included non-contrast-enhanced images, 18 contrast-enhanced images, in two contrast-related information was not specified and in two both contrast- and non-contrast-enhanced scans were included. Regarding tumor segmentation, a manual approach was followed in 48 studies, and automatic and semi-automatic segmentations were applied in two and 31 studies, respectively; three studies applied both methodologies (for verification or a different approach according to the imaging modality used) and five studies did not specify the method utilized for tumor segmentation.

View this table:
Table 1.

Methodological characteristics of the studies (N = 89) included in the systematic review. For those studies with the same name for the first author and published the same year, a hashtag was added to unequivocally indicate those that were included in the different meta-analyses and consequently, that are represented in the forest plots.

Clinical characteristics of the studies

The 89 studies evaluated in the qualitative synthesis included a total of 32,084 patients with NSCLC. Although most of the studies included >200 patients, in 42 publications, the sample size did not reach this figure and in 11 studies sample size was even lower than 100. All studies were retrospective and mostly unicentric (n = 72); the number of participant centers was not specified in one study19. In general, basic clinical and demographic information collected included sex, age, smoking status, TNM stage, histology, and treatment status at the moment of image acquisition, although this information was not available in 13, 9, 22, 34, 22 and 24 studies out of the 89 assessed, respectively. The clinical characteristics of the patients included in the 89 studies are depicted in Table 2. The median [range]/ mean ± standard deviation (SD) age of patients was 61.78 [59–64.17] years and 61.71 ± 3.64 years, respectively. In terms of sex, the total population was balanced, with 13,574 females and 14,066 males. The smoking history was available for 23,200 patients, and many were non-smokers (n = 12,813); while smoking history was unknown for 1,146 patients. Out of the 55 studies detailing information about the TNM stage, the majority of them (n = 40) included information about the four stages (I-IV), either provided per group or grouped in stages I-II and stages III-IV. Among the 15 studies that did not include patients of all stages, two studies included only early stage patients (stages I and II)20, 21, two included patients stage II-IV22, 23, six included only patients of stages III and IV24-29 (three of them with a majority of stage IV patients24, 27, 29), and five included patients of stages I-III without including the most advanced stage19, 30-33. A total of 65 studies included patients with adenocarcinoma: 43 exclusively including this histology subtype and 22 including other NSCLC histology types as well. Finally, in most of the cases (n = 65), images were acquired before patients received any treatment, with two studies also including post-treatment images34, 35. In 24 studies, no information on treatment was detailed, although in some of them image acquisition before surgery31, 36-41, before polymerase chain reaction (PCR)42, or before pathological diagnosis43 was detailed as an inclusion criterion. In five studies19, 44-47, authors specify that patients had not received radiotherapy or chemotherapy, but no information on targeted therapy was provided. Finally, only one study48 out of the 89 included in the systematic review, which did not meet the inclusion criteria to be considered for the meta-analysis, included patients who had received treatment with tyrosine kinase inhibitors (TKIs).

View this table:
Table 2.

Clinical characteristics of the studies (N = 89) included in the systematic review. For those studies with the same name for the first author and published the same year, a hashtag was added to unequivocally indicate those that were included in the different meta-analyses and consequently, that are represented in the forest plots.

Quantitative analysis (meta-analysis)

A total of 38 studies met the inclusion criteria for the quantitative assessment (n = 17,066 patients). Three main different meta-analyses including radiomics-based models were conducted: 1) a meta-analysis including studies focused on the detection of EGFR (n = 34 studies)17, 25-28, 32, 33, 36, 38, 39, 46, 49-71; 2) a meta-analysis including studies focused on the detection of ALK (n = 3 studies)72-74; 3) a meta-analysis including studies focused on the detection of KRAS (n = 4 studies)47, 50, 54, 62. In three studies, authors developed models for the detection of both EGFR and KRAS50, 54, 62. Furthermore, a separate meta-analysis was conducted for combined models (radiomics features + clinical variables) for the prediction of EGFR (not enough studies for ALK or KRAS mutations). Studies included in all the meta-analyses conducted are summarized in Supplementary Table S3. Details on the radiomics features included in the EGFR, ALK, and KRAS models are summarized in Supplementary Table S4, Supplementary Table S5 and Supplementary Table S6, respectively. In terms of radiomics variables, models grouped different combinations of first order, shape, gray level co-occurrence matrix (GLCM), gray level size zone matrix (GLSZM), gray level run length matrix (GLRLM), neighboring gray tone difference matrix (NGTDM), and gray level dependence matrix (GLDM) features. Clinical data included sex, smoking history, and/or histological type in the majority of studies.

EGFR

Results of the meta-analysis focused on models built with radiomics features are summarized in Figure 2. Note that this meta-analysis also included a study50 in which predictions were based on features extracted by a multi-channel and multi-task deep learning model with the ability to simultaneously detect EGFR and KRAS oncogene mutations; and consequently, did not include radiomics features (only single-task results for the independent prediction of EGFR and KRAS were considered for the quantitative analysis). A hierarchical sROC curve was plotted for the included 24 studies 17, 25, 27, 32, 36, 39, 46, 49-51, 54, 55, 57-60, 62, 64-66, 68-71 that evaluate the performance of AI algorithms in predicting EGFR mutation status in NSCLC (Supplementary Figure S1). Eight studies assessed more than one model32, 36, 39, 50, 59, 60, 65, 70. As observed, radiomics-based models exhibited high diagnostic performance in predicting EGFR mutation status with an overall AUC of 0.766. The AI algorithms’ sensitivity in determining the EGFR mutation status varied from 0.362 to 0.948, resulting in an estimate of 0.753 (95% CI 0.721–0.783). The FPR of these algorithms ranged from 0.022 to 0.761, with a estimate of 0.346 (95% CI 0.305–0.390). Detecting a positive case for EGFR mutation was almost six times more likely than not detecting it (DOR = 5.70 [95% CI 4.74–6.81]).

Figure 2.
Figure 2.

Forest plots of the included studies developing radiomics models using machine learning and/or deep learning methods for the prediction of EGFR mutation status. Numbers are estimated with 95% CIs in brackets and indicated by horizontal lines. For those studies with the same name for the first author and published the same year, a hashtag was added to unequivocally tag them as done in Tables 1 and 2 and in the reference list. EGFR, epidermal growth factor receptor; CI, confidence interval; DOR, diagnostic odds ration; FPR, false positive rate.

The effect of adding clinical variables to radiomics models or to models including both radiomics and deep features65 (models including clinical data and radiomic or deep features referred in this work as combined models) in the prediction of EGFR mutation was also analyzed. This meta-analysis included 23 studies25, 26, 28, 32, 33, 38, 39, 46, 51-53, 56-58, 61-69, of which four of them developed more than one model32, 39, 56, 67. Results are depicted in Figure 3 and sROC curve in Supplementary Figure S1. Overall, the performance of combined models slightly improved compared to radiomics models, with an AUC of 0.811 and a sensitivity of 0.800 (95% CI 0.767–0.830; model’s sensitivity ranging from 0.523 to 0.944). The FPR resulted similar with a value of 0.335 (95% CI 0.279–0.396; model’s FPR ranging from 0.167 to 0.760.). Detecting a positive case for EGFR mutation with combined models was more than eight times more likely than not detecting it (DOR = 8.35 [95% CI 6.77–10.20]).

Figure 3.
Figure 3.

Forest plots of the included studies developing combined models (radiomics + clinical data) using machine learning and/or deep learning methods for the prediction of EGFR mutation status. Numbers are estimates with 95% CIs in brackets and indicated by horizontal lines. For those studies with the same name for the first author and published the same year, a hashtag was added to unequivocally tag them as done in Tables 1 and 2 and in the reference list. EGFR, epidermal growth factor receptor; CI, confidence interval; DOR, diagnostic odds ration; FPR, false positive rate.

ALK

The meta-analysis focused on radiomics-based models included three studies72-74, one of which developed two different models, one based on pre-contrast images and another one on post-contrast images73. An overall AUC of 0.831 was obtained for the prediction of ALK aberration, with a sensitivity ranging from 0.682 to 0.825, resulting in an estimate of 0.754 (95% CI 0.639–0.841). The FPR of these algorithms ranged from 0.167 to 0.277, with an estimate of 0.225 (95% CI 0.163–0.302). Detecting a positive case for ALK aberration was 11 times more likely than not detecting it (DOR = 5.70 [95% CI 5.83–19.10]) (Figure 4 and Supplementary Figure S2). Given the lack of enough studies developing combined models, a meta-analysis to assess the effects of adding clinical variables in the prediction of ALK aberration was not possible. The only study74 that developed a model including age, sex, smoking history, smoking index, clinical stage, distal metastasis and pathological invasiveness of the tumor in combination with conventional CT features and different first order, GLCM, GLSZM, and GLRL radiomics features demonstrated increased performance in predicting ALK aberration of the combined model vs the radiomics-based model, but only in the primary cohort (AUC, 0.83–0.88, p = 0.01), not in the testing cohort (AUC, 0.80–0.88, p = 0.29).

Figure 4.
Figure 4.

Forest plots of the included studies developing radiomics models using machine learning and/or deep learning methods for the prediction of A) ALK and B) KRAS mutation status. Numbers are estimates with 95% CIs in brackets and indicated by horizontal lines. For those studies with the same name for the first author and published the same year, a hashtag was added to unequivocally tag them as done in Tables 1 and 2 and in the reference list. ALK, anaplastic lymphoma kinase; CI, confidence interval; DOR, diagnostic odds ration; FPR, false positive rate; KRAS, Kirsten rat sarcoma viral oncogene homologue.

KRAS

Four studies met the inclusion criteria for the meta-analysis assessing models for KRAS mutation prediction47, 50, 54, 62, among which, three of them also developed models for EGFR mutation prediction50, 54, 62. KRAS/EGFR models were independently built except in one study, in which a multi-channel multi-task DL model for the prediction of both KRAS and EGFR mutations was developed50. However, and according to the inclusion criteria, only single-task metrics were considered for the quantitative analysis despite the multi-channel version displayed the highest performance for the simultaneous detection of both oncogenic driver mutations. Results of the meta-analysis evaluating radiomics-based models are shown in Figure 4 and Supplementary Figure S3. KRAS mutation was predicted with an overall AUC of 0.732 and a sensitivity of 0.744 (95% CI 0.605–0.846; model’s sensitivity ranging from 0.641 to 0.875. The FPR was 0.376 (95% CI 0.274–0.491; model’s FPR ranging from 0.259 to 0.468). Detecting a positive case for KRAS mutation with radiomics-based models was more than five times more likely than not detecting it (DOR = 8.35 [95% CI 1.98–11.70]). Like ALK, the lack of enough KRAS studies made it impossible to perform a meta-analysis analyzing combined models. Only Ríos Velázquez et al.62 built a model including age, sex, smoking status, race, and clinical stage together with radiomics features that performed similar to the radiomics model (AUC = 0.69 [95% CI: 0.63–0.75] vs AUC = 0.63 [95% CI: 0.57–0.69]) and worse than a model developed only with clinical data AUC = 0.75 [95% CI: 0.69–0.80].

Meta-regression and subgroup analysis

The possible effects of different predictors on the predictive performance of the models was evaluated for EGFR mutation (not enough studies were available for ALK or KRAS mutations). Neither age, nor the use of contrast, nor the type of segmentation (manual/semi-automatic/automatic), nor the model (radiomics/combined), nor the AI methodology (machine learning/deep learning), yielded statistically significant results (Supplementary Table S7).

DISCUSSION

At present, molecular testing performed on biopsied tissue remains the gold standard for diagnosis and genotyping in advanced NSCLC75, 76. However, given the associated limitations and inconveniences, such as the lack of enough tissue for successful testing77, 78, or the long turnaround times76, there is a need to validate and incorporate new procedures into routine clinical practice. In recent years, liquid biopsy has emerged as a promising alternative in NSCLC, especially in clinical scenarios78. Likewise, radiomics have shown encouraging results in prognosis and prediction in this setting79. In general, both methodologies possess great potential, since they are both simple, straightforward to do, and repeatable at patient follow-up visits, which makes it possible to gather important data about the type of tumor, its aggressiveness, its progression, and its response to therapy80. Radiomics has the additional advantage of only requiring medical images and capturing patient-level and tissue-level heterogeneity, such as CT scans in lung cancer, that are usually acquired as part of the patient’s standard journey, representing an affordable methodology both in terms of resources and costs. It is important that new techniques are properly validated to facilitate their standardization, prior to incorporation into the routine clinical workflow.

To our knowledge, this is the first systematic review and meta-analysis that analyzes the performance and applicability of different imaging-based models for the prediction of three of the most common oncogene mutations—EGFR, ALK and KRAS—in NSCLC from a clinical perspective and with a special focus on AI methodologies. So far, results were only available for EGFR studies and did not take clinical aspects into account81. Thus, the results of our different meta-analyses demonstrate that AI-based models developed with CT-derived radiomics features showed good performance in predicting EGFR, ALK, and KRAS mutations with a sensitivity of 0.753 [95% CI (0.721–0.783)], 0.754 [95% CI (0.639–0.841)] and 0.744 [95% CI (0.605–0.846)], respectively. Whether the inclusion of clinical variables increase models’ performance cannot be concluded from our results, although we believe that increasing the number of studies would probably confirm the trends observed in our quantitative analysis of EGFR mutation.

Our outcomes point to radiomics as a candidate screening tool for oncogene mutation status determination. We especially focused on CT-based models, aiming to obtain conclusions as applicable as possible to the standard clinical workflow since CT remains the most utilized imaging tool in NSCLC82. From our work, we conclude that in addition to additional validation of our findings that future studies should be conducted that consider the following important aspects. Firstly, a minimum sample size should be guaranteed to ensure the reliability of the results obtained with AI-based models83, 84. In both our systematic review and meta-analyses, more than half of the studies were conducted in >200 patients (n = 46/89 and n = 23/38 [n = 20/34 for EGFR, n = 2/3 for ALK and n = 3/4 for KRAS]), but still a sizable number had small sample sizes, which definitely limited the relevance of their conclusions. Multicentric designs would be also desirable to get more solid conclusions, an approach that few studies followed (n = 16/89 in the systematic review and n = 10/39 in the meta-analysis [n = 10/34 for EGFR, n = 0/3 for ALK and n = 3/4 for KRAS]). Secondly, including independent cohorts for external validations would reinforce the results, leading to more robust and reproducible models. Out of the 89 studies included in the qualitative analysis, only 9 used external cohorts for validation43, 46, 60, 63, 67, 68, 85-87, of which five were included in the EGFR meta-analysis46, 60, 63, 67, 68. Finally, it is important that patient populations reflect clinical practice. Thus, considering the potential applicability of the models for diagnostic purposes, studies should be conducted in treatment naïve populations to avoid possible therapy-related confounding effects, an inclusion criterion mostly applied in the studies evaluated in this work, but still missing in some of them. Additionally, studies should be carried out preferably in stage III-IV NSCLC patients (especially in those at stage IV, for whom clinical guidelines recommend molecular testing75, 76). As demonstrated in this work, most of the studies published so far do not provide information on TNM stage or include patients from all stages. Despite the heterogeneity of the studies evaluated, we believe that the evidence provided is enough as to demonstrate the potential of radiomics in oncogene mutation status determination. Thus, AI-based models using radiomics extracted from CT scans could be effective non-invasive screening tools to detect targetable driver mutations in NSCLC with good sensitivity and moderate specificity. These tools would not be intended to replace gold standard techniques, such as PCR or next-generation sequencing, but to allow for the potential earlier identification of ideal candidates to be genetically tested, saving time, costs, and samples. Consequently, a high sensitivity would ensure the identification of oncogene mutation positive patients for whom laboratory-based testing would be subsequently confirmed.

When the influence of different factors on the prediction of EGFR mutation was evaluated, no statistically significant results were obtained, probably due to the limited number of studies included and the presence of missing data. However, some of those factors might play an essential role and should be considered when developing accurate models to be potentially implemented into clinical practice. Indeed, some of the studies included in our qualitative analysis analyzed the impact of different methodological aspects on the performance of the models. For example, Huang et al.35 demonstrated that interobserver variability in tumor segmentation affects the use of radiomics to predict oncogene mutation status, which suggests that automatic or semi-automatic models might be more suitable. In the study by Shiri et al.88, the application to radiomics features of ComBat harmonization improved the performance of the models toward more successful prediction of EGFR and KRAS mutations. Likewise, other authors have pointed to the impact of the experimental settings on the robustness of radiomics features89, or the influence of CT slice thickness on the predictive performance of radiomics-based models31. It is also worth mentioning the relevance of using a particular AI methodology. Although we found no differences in the EGFR mutation predictive performance between ML and DL methods, most likely due to the limited number of available DL-based studies, the latter might offer some advantages over the former. Thus, while in radiomics analysis a process of lesion segmentation and subsequent feature extraction is required, which introduces certain degree of variability and can be a high time-consuming task, DL models only required a bounding box of the lesion, greatly reducing this effect. On the other hand, DL models, and in particular end-to-end convolutional neural network (CNN) models, such those developed in most of the DL studies included in our work37, 42, 43, 50, 58, 68, 85, 86, 90, 91, are generally more complex in terms of the number of parameters, allowing to solve more complicated problems than traditional ML models. Considering available evidence, it seems reasonable to think that methodologic approaches should be carefully revised when validation studies are designed and conducted.

Our study has also some limitations, mainly derived from the limitations of the publications included. Thus, it is based on retrospective studies displaying great heterogeneity in terms of methodology and patient clinical characteristics, which clearly hamper the impact of our conclusions. Additionally, the limited available evidence for ALK and KRAS mutations, makes it difficult to draw solid conclusions. Despite this, our work gathers the most up-to-date and complete evidence (all models developed in each of the studies were analyzed) on imaging-based models for the prediction of three of the most important oncogene mutations in NSCLC, following a clinical approach and a special focus on AI models. Our exhaustive review and meta-analyses are intended to provide solid evidence for future research in the field.

In conclusion, radiomics-based models offer a useful and non-invasive method for determining the status of EGFR mutations in NSCLC and seem to retain similar predictive value for ALK and KRAS mutations. Additionally, although the inclusion of clinical variables tends to increase the performance of the models, further validation is required.

Data Availability

All data produced in the present work are contained in the manuscript.

CONFLICTS OF INTEREST

GJW reports personal fees from Quibim related to this work. He is a former employee of SOTIO Biotech Inc., and reports personal fees from Imaging Endpoints II, MiRanostics Consulting, Gossamer Bio, International Genomics Consortium, Angiex, Genomic Health, Oncacare, Rafael Pharmaceuticals, Roche, Immunocore, Kymera, and SPARC-all outside this submitted work; has ownership interest in MiRanostics Consulting, Exact Sciences, Moderna, Agenus, Aurinia Pharmaceuticals, and Circulogene-outside the submitted work; and has issued patents-all outside the submitted work. The remaining authors declare no conflicts of interest.

FUNDING

FIGURES

Supplementary Figure S1.
Supplementary Figure S1.

Hierarchical sROC curves of included studies for the comparative performance of radiomics models and combined models (radiomics + clinical data) using machine learning and/or deep learning methods for the prediction of EGFR mutation status (n = 24 and n = 23 studies, respectively). EGFR, epidermal growth factor receptor.

Supplementary Figure S2.
Supplementary Figure S2.

Hierarchical sROC curve of included studies for the performance of radiomics models for the prediction of ALK mutation status (n = 3). ALK, anaplastic lymphoma kinase.

Supplementary Figure S3.
Supplementary Figure S3.

Hierarchical sROC curve of included studies for the performance of radiomics models for the prediction of KRAS mutation status (n = 4). KRAS, Kirsten rat sarcoma viral oncogene homologue.

Supplementary tables

View this table:
Supplementary Table S1.

Search strategy applied for the qualitative analysis (systematic review).

View this table:
Supplementary Table S2.

Quality assessment results obtained after CLAIM evaluation.

View this table:
Supplementary Table S3.

Studies included in the different meta-analyses conducted.

View this table:
Supplementary Table S4.

Type of models (radiomic model/deep learning or combined [radiomic features + clinical variables]) developed in the studies for EGFR prediction and the radiomics/clinical features included. EGFR, epidermal growth factor receptor.

View this table:
Supplementary Table S5.

Type of models (radiomic model or combined [radiomic features + clinical variables]) developed in the studies for ALK prediction and the radiomics/clinical features included. ALK, anaplastic lymphoma kinase.

View this table:
Supplementary Table S6.

Type of models (radiomic model or combined [radiomic features + clinical variables]) developed in the studies for KRAS prediction and the radiomics/clinical features included. KRAS, Kirsten rat sarcoma viral oncogene homologue.

View this table:
Supplementary Table S7.

Results of the meta-regression analyzing the effects of age, type of segmentation (manual/semi-automatic/automatic), type of model (radiomics/combined [radiomic features + clinical data) and artificial intelligence methodology (machine learning/deep learning).

ACKNOWLEDGEMENTS

REFERENCES

  1. ↵
    Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 Cancers in 185 countries. CA Cancer J Clin. 2021; 71(3): 209-249.
    OpenUrlCrossRefPubMed
  2. ↵
    Osmani L, Askin F, Gabrielson E et al. Current WHO guidelines and the critical role of immunohistochemical markers in the subclassification of non-small cell lung carcinoma (NSCLC): moving from targeted therapy to immunotherapy. Semin Cancer Biol. 2018; 52(Pt 1): 103–109.
    OpenUrlCrossRefPubMed
  3. ↵
    König D, Savic Prince S, Rothschild SI. Targeted therapy in advanced and metastatic non-small cell lung cancer. An update on treatment of the most important actionable oncogenic driver alterations. Cancers (Basel). 2021; 13(4).
  4. ↵
    Chiu YW, Kao YH, Simoff MJ et al. Costs of biopsy and complications in patients with lung cancer. Clinicoecon Outcomes Res. 2021; 13: 191–200.
    OpenUrl
  5. Manicone M, Poggiana C, Facchinetti A et al. Critical issues in the clinical application of liquid biopsy in non-small cell lung cancer. J Thorac Dis. 2017; 9(Suppl 13): S1346–s1358.
    OpenUrl
  6. ↵
    Young M SR. Percutaneous lung lesion biopsy. [Updated 2023 Jun 19]. StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing.
  7. ↵
    Di Capua D, Bracken-Clarke D, Ronan K et al. The liquid biopsy for lung cancer: state of the art, limitations and future developments. Cancers (Basel). 2021; 13(16).
  8. ↵
    Wu G, Jochems A, Refaee T et al. Structural and functional radiomics for lung cancer. Eur J Nucl Med Mol Imaging. 2021; 48(12): 3961–3974.
    OpenUrl
  9. ↵
    Bera K, Braman N, Gupta A et al. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat Rev Clin Oncol. 2022; 19(2): 132–146.
    OpenUrlCrossRefPubMed
  10. ↵
    Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021; 372: n71.
    OpenUrlFREE Full Text
  11. ↵
    Mongan J, Moy L, Kahn CE. Checklist for artificial intelligence in medical imaging (CLAIM): a guide for authors and reviewers. Radiol Artif Intell. 2020; 2(2): e200029.
    OpenUrlCrossRefPubMed
  12. ↵
    Bossuyt PM, Reitsma JB. The STARD initiative. Lancet. 2003; 361(9351): 71.
    OpenUrlCrossRefPubMedWeb of Science
  13. Bossuyt PM, Reitsma JB, Bruns DE et al. Towards complete and accurate reporting of studies of diagnostic accuracy: the STARD initiative. Radiology. 2003; 226(1): 24–28.
    OpenUrlCrossRefPubMedWeb of Science
  14. Bossuyt PM, Reitsma JB, Bruns DE et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. Radiology. 2015; 277(3): 826-832.
    OpenUrlCrossRefPubMed
  15. ↵
    Cohen JF, Korevaar DA, Altman DG, et al. STARD 2015 guidelines for reporting diagnostic accuracy studies: explanation and elaboration. BMJ Open. 2016; 6(11): e012799.
    OpenUrlAbstract/FREE Full Text
  16. ↵
    Reitsma JB, Glas AS, Rutjes AW et al. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol. 2005; 58(10): 982–990.
    OpenUrlCrossRefPubMedWeb of Science
  17. ↵
    Nair JKR, Saeed UA, McDougall CC et al. Radiogenomic models using machine learning techniques to predict EGFR mutations in non-small cell lung cancer. Can Assoc Radiol J. 2021; 72(1): 109–119.
    OpenUrlPubMed
  18. ↵
    Shiri I, Maleki H, Hajianfar G et al. Next-generation radiogenomics sequencing for prediction of EGFR and KRAS mutation status in NSCLC patients using multimodal imaging and machine learning algorithms. Mol Imaging Biol. 2020; 22(4): 1132–1148.
    OpenUrlPubMed
  19. ↵
    Dang Y, Wang R, Qian K et al. Clinical and radiological predictors of epidermal growth factor receptor mutation in nonsmall cell lung cancer. J Appl Clin Med Phys. 2021; 22(1): 271–280.
    OpenUrl
  20. ↵
    Omura K, Murakami Y, Hashimoto K et al. Detection of EGFR mutations in early-stage lung adenocarcinoma by machine learning-based radiomics. Transl Cancer Res. 2023; 12(4): 837–847.
    OpenUrl
  21. ↵
    Wang X, Kong C, Xu W et al. Decoding tumor mutation burden and driver mutations in early stage lung adenocarcinoma using CT-based radiomics signature. Thorac Cancer. 2019; 10(10): 1904–1912.
    OpenUrl
  22. ↵
    Liu Q, Sun D, Li N et al. Predicting EGFR mutation subtypes in lung adenocarcinoma using (18)F-FDG PET/CT radiomic features. Transl Lung Cancer Res. 2020; 9(3): 549–562.
    OpenUrlCrossRef
  23. ↵
    Yang L, Xu P, Li M et al. PET/CT radiomic features: a potential biomarker for EGFR mutation status and survival outcome prediction in NSCLC patients treated with TKIs. Front Oncol. 2022; 12: 894323.
    OpenUrl
  24. ↵
    Hong D, Xu K, Zhang L et al. Radiomics signature as a predictive factor for EGFR mutations in advanced lung adenocarcinoma. Front Oncol. 2020; 10: 28.
    OpenUrlPubMed
  25. ↵
    Lu J, Ji X, Wang L et al. Machine learning-based radiomics for prediction of epidermal growth factor receptor mutations in lung adenocarcinoma. Dis Markers. 2022; 2022: 2056837.
    OpenUrl
  26. ↵
    Wu S, Shen G, Mao J et al. CT radiomics in predicting EGFR mutation in non-small cell lung cancer: a single institutional study. Front Oncol. 2020; 10: 542957.
    OpenUrl
  27. ↵
    Yang C, Chen W, Gong G et al. Application of CT radiomics features to predict the EGFR mutation status and therapeutic sensitivity to TKIs of advanced lung adenocarcinoma. Transl Cancer Res. 2020; 9(11): 6683–6690.
    OpenUrl
  28. ↵
    Zhang L, Chen B, Liu X et al. Quantitative biomarkers for prediction of epidermal growth factor receptor mutation in non-small cell lung cancer. Transl Oncol. 2018; 11(1): 94–101.
    OpenUrl
  29. ↵
    Zhu Y, Guo YB, Xu D et al. A computed tomography (CT)-derived radiomics approach for predicting primary co-mutations involving TP53 and epidermal growth factor receptor (EGFR) in patients with advanced lung adenocarcinomas (LUAD). Ann Transl Med. 2021; 9(7): 545.
    OpenUrl
  30. ↵
    Aide N, Weyts K, Lasnon C. Prediction of the presence of targetable molecular alteration(s) with clinico-metabolic (18) F-FDG PET radiomics in non-Asian lung adenocarcinoma patients. Diagnostics (Basel). 2022; 12(10).
  31. ↵
    Li Y, Lu L, Xiao M et al. CT slice thickness and convolution Kernel affect performance of a radiomic model for predicting egfr status in non-small cell lung cancer: a preliminary study. Sci Rep. 2018; 8(1): 17913.
    OpenUrlPubMed
  32. ↵
    Zhu H, Song Y, Huang Z et al. Accurate prediction of epidermal growth factor receptor mutation status in early-stage lung adenocarcinoma, using radiomics and clinical features. Asia Pac J Clin Oncol. 2022; 18(6): 586–594.
    OpenUrl
  33. ↵
    Dong Y, Jiang Z, Li C et al. Development and validation of novel radiomics-based nomograms for the prediction of EGFR mutations and Ki-67 proliferation index in non-small cell lung cancer. Quant Imaging Med Surg. 2022; 12(5): 2658–2671.
    OpenUrl
  34. ↵
    Aerts HJ, Grossmann P, Tan Y et al. Defining a adiomic response phenotype: a pilot study using targeted therapy in NSCLC. Sci Rep. 2016; 6: 33860.
    OpenUrlPubMed
  35. ↵
    Huang Q, Lu L, Dercle L et al. Interobserver variability in tumor contouring affects the use of radiomics to predict mutational status. J Med Imaging (Bellingham). 2018; 5(1): 011005.
    OpenUrl
  36. ↵
    Feng Y, Song F, Zhang P et al. Prediction of EGFR mutation status in non-small cell lung cancer based on ensemble learning. Front Pharmacol. 2022; 13: 897597.
    OpenUrl
  37. ↵
    Huang X, Sun Y, Tan M et al. Three-dimensional convolutional neural network-based prediction of epidermal growth factor receptor expression status in patients with non-small cell lung cancer. Front Oncol. 2022; 12: 772770.
    OpenUrl
  38. ↵
    Jia TY, Xiong JF, Li XY et al. Identifying EGFR mutations in lung adenocarcinoma by noninvasive imaging using radiomics features and random forest modeling. Eur Radiol. 2019; 29(9): 4742–4750.
    OpenUrl
  39. ↵
    Liu G, Xu Z, Ge Y et al. 3D radiomics predicts EGFR mutation, exon-19 deletion and exon-21 L858R mutation in lung adenocarcinoma. Transl Lung Cancer Res. 2020; 9(4): 1212–1224.
    OpenUrl
  40. Mei D, Luo Y, Wang Y et al. CT texture analysis of lung adenocarcinoma: can Radiomic features be surrogate biomarkers for EGFR mutation statuses. Cancer Imaging. 2018; 18(1): 52.
    OpenUrl
  41. ↵
    Zhang T, Xu Z, Liu G et al. Simultaneous identification of EGFR,KRAS,ERBB2, and TP53 mutations in patients with non-small cell lung cancer by machine learning-derived three-dimensional radiomics. Cancers (Basel). 2021; 13(8).
  42. ↵
    Song J, Ding C, Huang Q et al. Deep learning predicts epidermal growth factor receptor mutation subtypes in lung adenocarcinoma. Med Phys. 2021; 48(12): 7891–7899.
    OpenUrl
  43. ↵
    Wang C, Xu X, Shao J et al. Deep learning to predict EGFR mutation and PD-L1 expression status in non-small-cell lung cancer on computed tomography images. J Oncol. 2021; 2021: 5499385.
    OpenUrl
  44. ↵
    Huang W, Wang J, Wang H et al. PET/CT based egfr mutation status classification of NSCLC using deep learning features and radiomics features. Front Pharmacol. 2022; 13: 898529.
    OpenUrl
  45. Kawazoe Y, Shiinoki T, Fujimoto K et al. Investigation of the combination of intratumoral and peritumoral radiomic signatures for predicting epidermal growth factor receptor mutation in lung adenocarcinoma. J Appl Clin Med Phys. 2023; 24(6): e13980.
    OpenUrl
  46. ↵
    Lu X, Li M, Zhang H et al. A novel radiomic nomogram for predicting epidermal growth factor receptor mutation in peripheral lung adenocarcinoma. Phys Med Biol. 2020; 65(5): 055012.
    OpenUrl
  47. ↵
    Wang J, Lv X, Huang W et al. Establishment and optimization of radiomics algorithms for prediction of KRAS gene mutation by integration of NSCLC gene mutation mutual exclusion information. Front Pharmacol. 2022; 13: 862581.
    OpenUrl
  48. ↵
    Yang X, Fang C, Li C et al. Can CT radiomics detect acquired T790M mutation and predict prognosis in advanced lung adenocarcinoma with progression after first- or second-generation EGFR TKIs? Front Oncol. 2022; 12: 904983.
    OpenUrl
  49. ↵
    Chang C, Zhou S, Yu H et al. A clinically practical radiomics-clinical combined model based on PET/CT data and nomogram predicts EGFR mutation in lung adenocarcinoma. Eur Radiol. 2021; 31(8): 6259–6268.
    OpenUrl
  50. ↵
    Dong Y, Hou L, Yang W et al. Multi-channel multi-task deep learning for predicting EGFR and KRAS mutations of non-small cell lung cancer on CT images. Quant Imaging Med Surg. 2021; 11(6): 2354–2375.
    OpenUrl
  51. ↵
    Gao J, Niu R, Shi Y et al. The predictive value of [(18)F]FDG PET/CT radiomics combined with clinical features for EGFR mutation status in different clinical staging of lung adenocarcinoma. EJNMMI Res. 2023; 13(1): 26.
    OpenUrl
  52. Huo JW, Luo TY, Diao L et al. Using combined CT-clinical radiomics models to identify epidermal growth factor receptor mutation subtypes in lung adenocarcinoma. Front Oncol. 2022; 12: 846589.
    OpenUrl
  53. ↵
    Jiang M, Yang P, Li J et al. Computed tomography-based radiomics quantification predicts epidermal growth factor receptor mutation status and efficacy of first-line targeted therapy in lung adenocarcinoma. Front Oncol. 2022; 12: 985284.
    OpenUrl
  54. ↵
    Le NQK, Kha QH, Nguyen VH et al. Machine learning-based radiomics signatures for EGFR and KRAS mutations prediction in non-small-cell lung cancer. Int J Mol Sci. 2021; 22(17).
  55. ↵
    Li S, Li Y, Zhao M et al. Combination of (18)F-fluorodeoxyglucose PET/CT radiomics and clinical features for predicting epidermal growth factor receptor mutations in lung adenocarcinoma. Korean J Radiol. 2022; 23(9): 921–930.
    OpenUrl
  56. ↵
    Li S, Luo T, Ding C et al. Detailed identification of epidermal growth factor receptor mutations in lung adenocarcinoma: Combining radiomics with machine learning. Med Phys. 2020; 47(8): 3458–3466.
    OpenUrl
  57. ↵
    Li X, Yin G, Zhang Y et al. Predictive power of a radiomic signature based on (18)F-FDG PET/CT images for EGFR mutational status in NSCLC. Front Oncol. 2019; 9: 1062.
    OpenUrlCrossRefPubMed
  58. ↵
    Li XY, Xiong JF, Jia TY et al. Detection of epithelial growth factor receptor (EGFR) mutations on CT images of patients with lung adenocarcinoma using radiomics and/or multi-level residual convolutionary neural networks. J Thorac Dis. 2018; 10(12): 6624–6635.
    OpenUrl
  59. ↵
    Liu Y, Zhou J, Wu J et al. Development and validation of machine learning models to predict epidermal growth factor receptor mutation in non-small cell lung cancer: a multi-center retrospective radiomics study. Cancer Control. 2022; 29: 10732748221092926.
    OpenUrl
  60. ↵
    Ninomiya K, Arimura H, Chan WY et al. Robust radiogenomics approach to the identification of EGFR mutations among patients with NSCLC from three different countries using topologically invariant Betti numbers. PLoS One. 2021; 16(1): e0244354.
    OpenUrl
  61. ↵
    Ninomiya K, Arimura H, Tanaka K et al. Three-dimensional topological radiogenomics of epidermal growth factor receptor Del19 and L858R mutation subtypes on computed tomography images of lung cancer patients. Comput Methods Programs Biomed. 2023; 236: 107544.
    OpenUrl
  62. ↵
    Rios Velazquez E, Parmar C, Liu Y et al. Somatic mutations drive distinct imaging phenotypes in lung cancer. Cancer Res. 2017; 77(14): 3922–3930.
    OpenUrlAbstract/FREE Full Text
  63. ↵
    Rossi G, Barabino E, Fedeli A et al. Radiomic detection of EGFR mutations in NSCLC. Cancer Res. 2021; 81(3): 724–731.
    OpenUrlAbstract/FREE Full Text
  64. ↵
    Tu W, Sun G, Fan L et al. Radiomics signature: A potential and incremental predictor for EGFR mutation status in NSCLC patients, comparison with CT morphology. Lung Cancer. 2019; 132: 28–35.
    OpenUrl
  65. ↵
    Wang C, Ma J, Shao J et al. Predicting EGFR and PD-L1 status in NSCLC patients using multitask ai system based on CT images. Front Immunol. 2022; 13: 813072.
    OpenUrl
  66. ↵
    Weng Q, Hui J, Wang H et al. Radiomic feature-based nomogram: a novel technique to predict EGFR-activating mutations for EGFR tyrosin kinase inhibitor therapy. Front Oncol. 2021; 11: 590937.
    OpenUrl
  67. ↵
    Yang X, Liu M, Ren Y et al. Using contrast-enhanced CT and non-contrast-enhanced CT to predict EGFR mutation status in NSCLC patients-a radiomics nomogram analysis. Eur Radiol. 2022; 32(4): 2693–2703.
    OpenUrl
  68. ↵
    Zhang B, Qi S, Pan X et al. Deep CNN model using CT radiomics feature mapping recognizes EGFR gene mutation status of lung adenocarcinoma. Front Oncol. 2020; 10: 598721.
    OpenUrl
  69. ↵
    Zhang G, Cao Y, Zhang J et al. Predicting EGFR mutation status in lung adenocarcinoma: development and validation of a computed tomography-based radiomics signature. Am J Cancer Res. 2021; 11(2): 546–560.
    OpenUrl
  70. ↵
    Zhang M, Bao Y, Rui W et al. Performance of (18)F-FDG PET/CT radiomics for predicting EGFR mutation status in patients with non-small cell lung cancer. Front Oncol. 2020; 10: 568857.
    OpenUrl
  71. ↵
    Zhao HY, Su YX, Zhang LH et al. Prediction model based on 18F-FDG PET/CT radiomic features and clinical factors of EGFR mutations in lung adenocarcinoma. Neoplasma. 2022; 69(1): 233–241.
    OpenUrl
  72. ↵
    Chang C, Sun X, Wang G et al. A machine learning model based on PET/CT radiomics and clinical characteristics predicts ALK rearrangement status in lung adenocarcinoma. Front Oncol. 2021; 11: 603882.
    OpenUrl
  73. ↵
    Ma DN, Gao XY, Dan YB et al. Evaluating solid lung adenocarcinoma anaplastic lymphoma kinase gene rearrangement using noninvasive radiomics biomarkers. Onco Targets Ther. 2020; 13: 6927–6935.
    OpenUrl
  74. ↵
    Song L, Zhu Z, Mao L et al. Clinical, conventional ct and radiomic feature-based machine learning models for predicting ALK rearrangement status in lung adenocarcinoma patients. Front Oncol. 2020; 10: 369.
    OpenUrl
  75. ↵
    Hendriks LE, Kerr KM, Menis J et al. Oncogene-addicted metastatic non-small-cell lung cancer: ESMO clinical practice guideline for diagnosis, treatment and follow-up. Ann Oncol. 2023; 34(4): 339–357.
    OpenUrl
  76. ↵
    Ettinger DS, Wood DE, Aisner DL et al. Non-Small Cell Lung Cancer, Version 3.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw. 2022; 20(5): 497-530.
    OpenUrlCrossRefPubMed
  77. ↵
    Ferry-Galow KV, Datta V, Makhlouf HR et al. What can be done to improve research biopsy quality in oncology clinical trials? J Oncol Pract. 2018; 14(11): Jop1800092.
    OpenUrl
  78. ↵
    Gutierrez ME, Choi K, Lanman RB et al. Genomic profiling of advanced non-small cell lung cancer in community settings: gaps and opportunities. Clin Lung Cancer. 2017; 18(6): 651–659.
    OpenUrlCrossRefPubMed
  79. ↵
    Fornacon-Wood I, Faivre-Finn C, O’Connor JPB et al. Radiomics as a personalized medicine tool in lung cancer: separating the hope from the hype. Lung Cancer. 2020; 146: 197–208.
    OpenUrlCrossRefPubMed
  80. ↵
    Neri E, Del Re M, Paiar F et al. Radiomics and liquid biopsy in oncology: the holons of systems medicine. Insights Imaging. 2018; 9(6): 915–924.
    OpenUrlPubMed
  81. ↵
    Nguyen HS, Ho DKN, Nguyen NN et al. Predicting EGFR mutation status in non-small cell lung cancer using artificial intelligence: a systematic review and meta-analysis. Acad Radiol. 2023.
  82. ↵
    Steinert HC. PET and PET-CT of lung cancer. Methods Mol Biol. 2011; 727: 33–51.
    OpenUrlPubMed
  83. ↵
    Alwosheel A, van Cranenburgh S, Chorus CG. Is your dataset big enough? Sample size requirements when using artificial neural networks for discrete choice analysis. J Choice Model. 2018; 28: 167–182.
    OpenUrl
  84. ↵
    Rajput D, Wang W-J, Chen C-C. Evaluation of a decided sample size in machine learning applications. BMC Bioinformatics. 2023; 24(1): 48.
    OpenUrlCrossRef
  85. ↵
    Mu W, Jiang L, Zhang J et al. Non-invasive decision support for NSCLC treatment using PET/CT radiomics. Nat Commun. 2020; 11(1): 5228.
    OpenUrlCrossRefPubMed
  86. ↵
    Zhao W, Yang J, Ni B et al. Toward automatic prediction of EGFR mutation status in pulmonary adenocarcinoma with 3D deep learning. Cancer Med. 2019; 8(7): 3532–3543.
    OpenUrl
  87. ↵
    Zuo Y, Liu Q, Li N et al. Optimal (18)F-FDG PET/CT radiomics model development for predicting EGFR mutation status and prognosis in lung adenocarcinoma: a multicentric study. Front Oncol. 2023; 13: 1173355.
    OpenUrl
  88. ↵
    Shiri I, Amini M, Nazari M et al. Impact of feature harmonization on radiogenomics analysis: prediction of EGFR and KRAS mutations from non-small cell lung cancer PET/CT images. Comput Biol Med. 2022; 142: 105230.
    OpenUrl
  89. ↵
    Yip SSF, Parmar C, Kim J et al. Impact of experimental design on PET radiomics in predicting somatic mutation status. Eur J Radiol. 2017; 97: 8–15.
    OpenUrl
  90. ↵
    Shao J, Ma J, Zhang S et al. Radiogenomic system for non-invasive identification of multiple actionable mutations and PD-L1 expression in non-small cell lung cancer based on ct images. Cancers (Basel). 2022; 14(19).
  91. ↵
    Xiao Z, Cai H, Wang Y et al. Deep learning for predicting epidermal growth factor receptor mutations of non-small cell lung cancer on PET/CT images. Quant Imaging Med Surg. 2023; 13(3): 1286–1299.
    OpenUrl
  92. Agüloğlu N, Aksu A, Akyol M et al. Importance of pretreatment 18F-FDG PET/CT texture analysis in predicting EGFR and ALK mutation in patients with non-small cell lung cancer. Nuklearmedizin. 2022; 61(6): 433–439.
    OpenUrl
  93. Agazzi GM, Ravanelli M, Roca E et al. CT texture analysis for prediction of EGFR mutational status and ALK rearrangement in patients with non-small cell lung cancer. Radiol Med. 2021; 126(6): 786–794.
    OpenUrl
  94. Chen W, Hua Y, Mao D et al. A computed tomography-derived radiomics approach for predicting uncommon EGFR mutation in patients with NSCLC. Front Oncol. 2021; 11: 722106.
    OpenUrl
  95. Chen Q, Li Y, Cheng Q et al. EGFR mutation status and subtypes predicted by CT-based 3D radiomic features in lung adenocarcinoma. Onco Targets Ther. 2022; 15: 597–608.
    OpenUrl
  96. Choe J, Lee SM, Kim W et al. CT radiomics-based prediction of anaplastic lymphoma kinase and epidermal growth factor receptor mutations in lung adenocarcinoma. Eur J Radiol. 2021; 139: 109710.
    OpenUrl
  97. Digumarthy SR, Padole AM, Gullo RL et al. Can CT radiomic analysis in NSCLC predict histology and EGFR mutation status? Medicine (Baltimore). 2019; 98(1): e13963.
    OpenUrl
  98. Hao P, Deng BY, Huang CT et al. Predicting anaplastic lymphoma kinase rearrangement status in patients with non-small cell lung cancer using a machine learning algorithm that combines clinical features and CT images. Front Oncol. 2022; 12: 994285.
    OpenUrl
  99. He R, Yang X, Li T et al. A machine learning-based predictive model of epidermal growth factor mutations in lung adenocarcinomas. Cancers (Basel). 2022; 14(19).
  100. Hou D, Li W, Wang S et al. Different clinicopathologic and computed tomography imaging characteristics of primary and acquired EGFR T790M mutations in patients with non-small-cell lung cancer. Cancer Manag Res. 2021; 13: 6389–6401.
    OpenUrl
  101. Jiang M, Zhang Y, Xu J et al. Assessing EGFR gene mutation status in non-small cell lung cancer with imaging features from PET/CT. Nucl Med Commun. 2019; 40(8): 842–849.
    OpenUrl
  102. Kawazoe Y, Shiinoki T, Fujimoto K et al. Comparison of the radiomics-based predictive models using machine learning and nomogram for epidermal growth factor receptor mutation status and subtypes in lung adenocarcinoma. Phys Eng Sci Med. 2023; 46(1): 395–403.
    OpenUrl
  103. Koyasu S, Nishio M, Isoda H et al. Usefulness of gradient tree boosting for predicting histological subtype and EGFR mutation status of non-small cell lung cancer on (18)F FDG-PET/CT. Ann Nucl Med. 2020; 34(1): 49–57.
    OpenUrlCrossRef
  104. Li S, Ding C, Zhang H et al. Radiomics for the prediction of EGFR mutation subtypes in non-small cell lung cancer. Med Phys. 2019; 46(10): 4545–4552.
    OpenUrl
  105. Li H, Gao C, Sun Y et al. Radiomics analysis to enhance precise identification of epidermal growth factor receptor mutation based on positron emission tomography images of lung cancer patients. J Biomed Nanotechnol. 2021; 17(4): 691–702.
    OpenUrl
  106. Liu Y, Kim J, Balagurunathan Y et al. Radiomic features are associated with EGFR mutation status in lung adenocarcinomas. Clin Lung Cancer. 2016; 17(5): 441–448.e446.
    OpenUrlCrossRefPubMed
  107. Liu Z, Zhang T, Lin L et al. Applications of radiomics-based analysis pipeline for predicting epidermal growth factor receptor mutation status. Biomed Eng Online. 2023; 22(1): 17.
    OpenUrl
  108. Lu L, Sun SH, Yang H et al. Radiomics prediction of EGFR status in lung cancer-our experience in using multiple feature extractors and The Cancer Imaging Archive Data. Tomography. 2020; 6(2): 223–230.
    OpenUrl
  109. Ruan D, Fang J, Teng X. Efficient 18F-Fluorodeoxyglucose positron emission tomography/computed tomography-based machine learning model for predicting epidermal growth factor receptor mutations in non-small cell lung cancer. Q J Nucl Med Mol Imaging. 2022.
  110. Trivizakis E, Souglakos J, Karantanas A et al. Deep radiotranscriptomics of non-small cell lung carcinoma for assessing molecular and histology subtypes with a data-driven analysis. Diagnostics (Basel). 2021; 11(12).
  111. Yamazaki M, Yagi T, Tominaga M et al. Role of intratumoral and peritumoral CT radiomics for the prediction of EGFR gene mutation in primary lung cancer. Br J Radiol. 2022; 95(1140): 20220374.
    OpenUrl
  112. Yang B, Ji HS, Zhou CS et al. (18)F-fluorodeoxyglucose positron emission tomography/computed tomography-based radiomic features for prediction of epidermal growth factor receptor mutation status and prognosis in patients with lung adenocarcinoma. Transl Lung Cancer Res. 2020; 9(3): 563–574.
    OpenUrl
  113. Zhang J, Zhao X, Zhao Y et al. Value of pre-therapy (18)F-FDG PET/CT radiomics in predicting EGFR mutation status in patients with non-small cell lung cancer. Eur J Nucl Med Mol Imaging. 2020; 47(5): 1137–1146.
    OpenUrlPubMed
  114. Zhang T, Liu Z, Lin L et al. Detection of the gene mutation of epidermal growth factor receptor in lung adenocarcinoma by radiomic features from a small amount of PET data. Nucl Med Commun. 2023.
  115. Zhao W, Wu Y, Xu Y et al. The potential of radiomics nomogram in non-invasively prediction of epidermal growth factor receptor mutation status and subtypes in lung adenocarcinoma. Front Oncol. 2019; 9: 1485.
    OpenUrl

REFERENCES

  1. Chang C, Zhou S, Yu H et al. A clinically practical radiomics-clinical combined model based on PET/CT data and nomogram predicts EGFR mutation in lung adenocarcinoma. Eur Radiol. 2021; 31(8): 6259–6268.
    OpenUrl
  2. Chang C, Sun X, Wang G et al. A machine learning model based on PET/CT radiomics and clinical characteristics predicts ALK rearrangement status in lung adenocarcinoma. Front Oncol. 2021; 11: 603882.
    OpenUrl
  3. Dong Y, Hou L, Yang W et al. Multi-channel multi-task deep learning for predicting EGFR and KRAS mutations of non-small cell lung cancer on CT images. Quant Imaging Med Surg. 2021; 11(6): 2354–2375.
    OpenUrl
  4. Dong Y, Jiang Z, Li C et al. Development and validation of novel radiomics-based nomograms for the prediction of EGFR mutations and Ki-67 proliferation index in non-small cell lung cancer. Quant Imaging Med Surg. 2022; 12(5): 2658–2671.
    OpenUrl
  5. Feng Y, Song F, Zhang P et al. Prediction of EGFR mutation status in non-small cell lung cancer based on ensemble learning. Front Pharmacol. 2022; 13: 897597.
    OpenUrl
  6. Gao J, Niu R, Shi Y et al. The predictive value of [(18)F]FDG PET/CT radiomics combined with clinical features for EGFR mutation status in different clinical staging of lung adenocarcinoma. EJNMMI Res. 2023; 13(1): 26.
    OpenUrl
  7. Huo JW, Luo TY, Diao L et al. Using combined CT-clinical radiomics models to identify epidermal growth factor receptor mutation subtypes in lung adenocarcinoma. Front Oncol. 2022; 12: 846589.
    OpenUrl
  8. Jia TY, Xiong JF, Li XY et al. Identifying EGFR mutations in lung adenocarcinoma by noninvasive imaging using radiomics features and random forest modeling. Eur Radiol. 2019; 29(9): 4742–4750.
    OpenUrl
  9. Jiang M, Yang P, Li J et al. Computed tomography-based radiomics quantification predicts epidermal growth factor receptor mutation status and efficacy of first-line targeted therapy in lung adenocarcinoma. Front Oncol. 2022; 12: 985284.
    OpenUrl
  10. Le NQK, Kha QH, Nguyen VH et al. Machine learning-based radiomics signatures for EGFR and KRAS mutations prediction in non-small-cell lung cancer. Int J Mol Sci. 2021; 22(17).
  11. Li XY, Xiong JF, Jia TY et al. Detection of epithelial growth factor receptor (EGFR) mutations on CT images of patients with lung adenocarcinoma using radiomics and/or multi-level residual convolutionary neural networks. J Thorac Dis. 2018; 10(12): 6624–6635.
    OpenUrl
  12. Li X, Yin G, Zhang Y et al. Predictive power of a radiomic signature based on (18)F-FDG PET/CT images for EGFR mutational status in NSCLC. Front Oncol. 2019; 9: 1062.
    OpenUrlCrossRefPubMed
  13. Li S, Luo T, Ding C et al. Detailed identification of epidermal growth factor receptor mutations in lung adenocarcinoma: Combining radiomics with machine learning. Med Phys. 2020; 47(8): 3458–3466.
    OpenUrl
  14. Li S, Li Y, Zhao M et al. Combination of (18)F-fluorodeoxyglucose PET/CT radiomics and clinical features for predicting epidermal growth factor receptor mutations in lung adenocarcinoma. Korean J Radiol. 2022; 23(9): 921–930.
    OpenUrl
  15. Liu G, Xu Z, Ge Y et al. 3D radiomics predicts EGFR mutation, exon-19 deletion and exon-21 L858R mutation in lung adenocarcinoma. Transl Lung Cancer Res. 2020; 9(4): 1212–1224.
    OpenUrl
  16. Liu Y, Zhou J, Wu J et al. Development and validation of machine learning models to predict epidermal growth factor receptor mutation in non-small cell lung cancer: a multi-center retrospective radiomics study. Cancer Control. 2022; 29: 10732748221092926.
    OpenUrl
  17. Lu X, Li M, Zhang H et al. A novel radiomic nomogram for predicting epidermal growth factor receptor mutation in peripheral lung adenocarcinoma. Phys Med Biol. 2020; 65(5): 055012.
    OpenUrl
  18. Lu J, Ji X, Wang L et al. Machine learning-based radiomics for prediction of epidermal growth factor receptor mutations in lung adenocarcinoma. Dis Markers. 2022; 2022: 2056837.
    OpenUrl
  19. Ma DN, Gao XY, Dan YB et al. Evaluating solid lung adenocarcinoma anaplastic lymphoma kinase gene rearrangement using noninvasive radiomics biomarkers. Onco Targets Ther. 2020; 13: 6927–6935.
    OpenUrl
  20. Nair JKR, Saeed UA, McDougall CC et al. Radiogenomic models using machine learning techniques to predict EGFR mutations in non-small cell lung cancer. Can Assoc Radiol J. 2021; 72(1): 109–119.
    OpenUrlPubMed
  21. Ninomiya K, Arimura H, Chan WY et al. Robust radiogenomics approach to the identification of EGFR mutations among patients with NSCLC from three different countries using topologically invariant Betti numbers. PLoS One. 2021; 16(1): e0244354.
    OpenUrl
  22. Ninomiya K, Arimura H, Tanaka K et al. Three-dimensional topological radiogenomics of epidermal growth factor receptor Del19 and L858R mutation subtypes on computed tomography images of lung cancer patients. Comput Methods Programs Biomed. 2023; 236: 107544.
    OpenUrl
  23. Rios Velazquez E, Parmar C, Liu Y et al. Somatic mutations drive distinct imaging phenotypes in lung cancer. Cancer Res. 2017; 77(14): 3922–3930.
    OpenUrlAbstract/FREE Full Text
  24. Rossi G, Barabino E, Fedeli A et al. Radiomic detection of EGFR mutations in NSCLC. Cancer Res. 2021; 81(3): 724–731.
    OpenUrlAbstract/FREE Full Text
  25. Song L, Zhu Z, Mao L et al. Clinical, conventional ct and radiomic feature-based machine learning models for predicting ALK rearrangement status in lung adenocarcinoma patients. Front Oncol. 2020; 10: 369.
    OpenUrl
  26. Tu W, Sun G, Fan L et al. Radiomics signature: A potential and incremental predictor for EGFR mutation status in NSCLC patients, comparison with CT morphology. Lung Cancer. 2019; 132: 28–35.
    OpenUrl
  27. Wang C, Ma J, Shao J et al. Predicting EGFR and PD-L1 status in NSCLC patients using multitask ai system based on CT images. Front Immunol. 2022; 13: 813072.
    OpenUrl
  28. Wang J, Lv X, Huang W et al. Establishment and optimization of radiomics algorithms for prediction of KRAS gene mutation by integration of NSCLC gene mutation mutual exclusion information. Front Pharmacol. 2022; 13: 862581.
    OpenUrl
  29. Weng Q, Hui J, Wang H et al. Radiomic feature-based nomogram: a novel technique to predict EGFR-activating mutations for EGFR tyrosin kinase inhibitor therapy. Front Oncol. 2021; 11: 590937.
    OpenUrl
  30. Wu S, Shen G, Mao J et al. CT radiomics in predicting EGFR mutation in non-small cell lung cancer: a single institutional study. Front Oncol. 2020; 10: 542957.
    OpenUrl
  31. Yang C, Chen W, Gong G et al. Application of CT radiomics features to predict the EGFR mutation status and therapeutic sensitivity to TKIs of advanced lung adenocarcinoma. Transl Cancer Res. 2020; 9(11): 6683–6690.
    OpenUrl
  32. Yang X, Liu M, Ren Y et al. Using contrast-enhanced CT and non-contrast-enhanced CT to predict EGFR mutation status in NSCLC patients-a radiomics nomogram analysis. Eur Radiol. 2022; 32(4): 2693–2703.
    OpenUrl
  33. Zhang L, Chen B, Liu X et al. Quantitative biomarkers for prediction of epidermal growth factor receptor mutation in non-small cell lung cancer. Transl Oncol. 2018; 11(1): 94–101.
    OpenUrl
  34. Zhang M, Bao Y, Rui W et al. Performance of (18)F-FDG PET/CT radiomics for predicting EGFR mutation status in patients with non-small cell lung cancer. Front Oncol. 2020; 10: 568857.
    OpenUrl
  35. Zhang B, Qi S, Pan X et al. Deep CNN model using CT radiomics feature mapping recognizes EGFR gene mutation status of lung adenocarcinoma. Front Oncol. 2020; 10: 598721.
    OpenUrl
  36. Zhang G, Cao Y, Zhang J et al. Predicting EGFR mutation status in lung adenocarcinoma: development and validation of a computed tomography-based radiomics signature. Am J Cancer Res. 2021; 11(2): 546–560.
    OpenUrl
  37. Zhao HY, Su YX, Zhang LH et al. Prediction model based on 18F-FDG PET/CT radiomic features and clinical factors of EGFR mutations in lung adenocarcinoma. Neoplasma. 2022; 69(1): 233–241.
    OpenUrl
  38. Zhu H, Song Y, Huang Z et al. Accurate prediction of epidermal growth factor receptor mutation status in early-stage lung adenocarcinoma, using radiomics and clinical features. Asia Pac J Clin Oncol. 2022; 18(6): 586–594.
    OpenUrl
Back to top
Posted May 31, 2024.
Download PDF
Data/Code
Email
Prediction of oncogene mutation status in non-small cell lung cancer: A systematic review and meta-analysis with a special focus on artificial-intelligence-based methods
Almudena Fuster-Matanzo, Alfonso Picó Peris, Fuensanta Bellvís Bataller, Ana Jimenez-Pastor, Glen J. Weiss, Luis Martí-Bonmatí, Antonio Lázaro Sánchez, Giuseppe L. Banna, Alfredo Addeo, Ángel Alberich-Bayarri
medRxiv 2024.05.31.24308261; doi: https://doi.org/10.1101/2024.05.31.24308261
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area