pith. sign in

arxiv: 2603.24922 · v2 · submitted 2026-03-26 · ⚛️ physics.med-ph

Robust Multicenter CT Radiogenomics for Dual EGFR and KRAS Prediction in Lung Cancer with Stability-Aware Modeling and SHAP Interpretation

Somayeh Sadat Mehrnia , Fatemeh Razavi , Helia Abedini , Niloofar Rahimi , Arman Rahmim , Mohammad Salmanpour This is my paper

Pith reviewed 2026-05-15 00:33 UTC · model grok-4.3

classification ⚛️ physics.med-ph
keywords CT radiomicsEGFR mutationKRAS mutationlung cancermulticenter validationsemi-supervised learningSHAP interpretabilityhandcrafted features
0
0 comments X

The pith

Handcrafted radiomics features from CT scans predict EGFR and KRAS mutations in lung cancer more reliably across centers than deep features.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper benchmarks handcrafted radiomics features against deep features and their fusion for three-class prediction of wild-type, KRAS-mutant, and EGFR-mutant status in non-small cell lung cancer. It assembles 1,023 CT scans from 12 public datasets spanning more than 20 centers, with 136 cases carrying mutation labels, and applies semi-supervised pseudo-labeling to the rest before external testing. Handcrafted features achieve an external AUC of 0.77 and accuracy of 0.77 while deep features fall to roughly 0.57, and SHAP analysis flags morphology and heterogeneity measures as the dominant predictors. A noninvasive CT route to mutation status could reduce reliance on tissue biopsies for guiding targeted therapies.

Core claim

Standardized handcrafted radiomics features within a multicenter semi-supervised framework provide a generalizable and interpretable approach for CT-based EGFR/KRAS stratification, as shown by superior external validation performance over deep feature representations.

What carries the argument

IBSI-compliant handcrafted radiomics features extracted after standardized preprocessing and deployed in stability-aware modeling for three-class mutation classification.

If this is right

  • HRF models maintain performance when moving from cross-validation to external multicenter testing while DFR models degrade.
  • SHAP analysis consistently identifies morphology- and heterogeneity-related radiomic phenotypes as the strongest predictors.
  • Fusion of HRF and DFR improves robustness over DFR alone but does not consistently exceed HRF performance.
  • Semi-supervised pseudo-labeling on unlabeled CT scans enables effective training when labeled mutation data remain scarce.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the HRF pipeline holds up in prospective trials it could serve as an initial noninvasive screen to triage patients away from immediate biopsy.
  • The same stability-aware extraction and modeling steps could be reused for other genetic markers or solid-tumor types where multicenter CT data exist.
  • The observed advantage of handcrafted over learned features suggests that explicit shape and texture descriptors may capture mutation-linked biology more stably than end-to-end networks on modest labeled sets.

Load-bearing premise

The 136 labeled patients drawn from 12 public datasets represent real-world tumor heterogeneity and semi-supervised pseudo-labeling adds no systematic bias to the unlabeled scans.

What would settle it

Apply the identical HRF pipeline to an independent multicenter cohort with fresh ground-truth mutation labels and check whether external AUC remains near 0.77 or falls substantially below it.

Figures

Figures reproduced from arXiv: 2603.24922 by Arman Rahmim, Fatemeh Razavi, Helia Abedini, Mohammad Salmanpour, Niloofar Rahimi, Somayeh Sadat Mehrnia.

Figure 1
Figure 1. Figure 1: Overall study workflow. Lung cancer CT images from 12 centers were converted from DICOM to NIfTI, reviewed by two experts, and preprocessed by clipping and normalization before handcrafted and deep radiomic feature extraction using PySERA. The extracted features were then split for external and cross-validation, normalized with min-max scaling, and harmonized using ComBat, while KRAS, EGFR, and wild-type l… view at source ↗
Figure 2
Figure 2. Figure 2: Cross-validation and external test performance of the top five HRF-based DRA+CA models under the SSL framework. Composite ranking was based exclusively on 5-fold cross-validation (CV) metrics from the model development cohort, including Accuracy, Precision, Recall, F1-score, and AUC, while external test (Ext) results are reported separately to assess generalizability. DRA: Dimension Reduction Algorithms; R… view at source ↗
Figure 3
Figure 3. Figure 3: Cross-validation and external test performance of the top five DRF-based DRA+CA models under the SSL framework. Composite ranking was based exclusively on 5-fold cross-validation (CV) metrics from the model development cohort, including Accuracy, Precision, Recall, F1-score, and AUC, while external test (Ext) results are reported separately to assess generalizability. DRA: Dimension Reduction Algorithms; L… view at source ↗
Figure 4
Figure 4. Figure 4: summarizes the top five DRA+CA combinations for the fused HRF+DRF dataset. DAE + HGB achieved the highest overall score (0.977) and the strongest internal performance, with CV Accuracy = 0.91 ± 0.00, Precision = 0.91 ± 0.00, Recall = 0.90 ± 0.00, and AUC = 0.96 ± 0.01. On external testing, this model maintained relatively favorable performance, with Accuracy = 0.65 ± 0.04, Precision = 0.65 ± 0.04, Recall =… view at source ↗
read the original abstract

Accurate identification of EGFR and KRAS mutations is essential for precision therapy in non-small cell lung cancer (NSCLC), but tissue genotyping is invasive and may not capture tumor heterogeneity. CT-based radiogenomics offers a noninvasive alternative, although generalization across centers remains challenging. We benchmarked handcrafted radiomics features (HRF), deep feature representations (DFR), and their fusion for three-class mutation prediction (wild-type, KRAS-mutant, and EGFR-mutant) with external testing. We curated 1,023 thoracic CT scans from 12 public datasets across more than 20 centers, including 136 patients with EGFR/KRAS labels. IBSI-compliant HRFs were extracted with standardized preprocessing, and DFRs were derived using PySERA. HRF-only, DFR-only, and fused HRF+DFR pipelines were evaluated using five-fold cross-validation and external testing. A semi-supervised pseudo-labeling strategy leveraged unlabeled CT scans, and SHAP supported interpretability. In external testing, HRF-based models generalized best, achieving AUC 0.77 +/- 0.07 and accuracy 0.77 +/- 0.00. DFR-based models showed a larger drop from cross-validation to external testing, with best external AUC around 0.57 +/- 0.05. Fusion improved robustness over DFR-only models but did not consistently outperform HRFs. SHAP identified morphology- and heterogeneity-related radiomic phenotypes as key predictors. Standardized handcrafted radiomics within a multicenter semi-supervised framework may provide a generalizable and interpretable approach for CT-based EGFR/KRAS stratification.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The manuscript benchmarks handcrafted radiomics features (HRF), deep feature representations (DFR), and their fusion for three-class prediction of EGFR/KRAS/wild-type status in NSCLC from 1,023 CT scans across 12 public datasets (>20 centers). Using 136 labeled cases and semi-supervised pseudo-labeling on the remaining ~887 scans, the authors report five-fold CV plus external testing, with HRF models achieving the best external performance (AUC 0.77 ± 0.07, accuracy 0.77 ± 0.00). SHAP is used for interpretability, and IBSI-compliant preprocessing is emphasized.

Significance. If the external generalization claim survives scrutiny of the pseudo-labeling step, the work would strengthen the case for standardized, interpretable CT radiogenomics as a noninvasive alternative to tissue genotyping. The multicenter scale, explicit comparison of feature families, and use of SHAP are positive elements. However, the small labeled cohort (136 patients) and dependence on pseudo-labels limit the strength of the robustness conclusion relative to prior single-center radiomics studies.

major comments (3)
  1. [Methods (semi-supervised pipeline)] Methods section on semi-supervised pseudo-labeling: the threshold, iteration count, and any quality check against the 136 ground-truth labels are not specified. Because the headline external AUC of 0.77 is obtained only after training on the pseudo-labeled pool, an ablation that reports performance when the model is trained solely on the 136 labeled cases is required to demonstrate that pseudo-labeling does not inject center-correlated label noise.
  2. [Results (external testing)] Results, external-testing paragraph and associated table: class prevalence in the external test set is not stated, nor is any strategy for three-class imbalance (e.g., class weights, oversampling). The reported accuracy of 0.77 ± 0.00 with zero variance is difficult to interpret without this information and raises the possibility that the metric is dominated by the majority class.
  3. [Results (model comparison)] Results, model-comparison subsection: no statistical test (e.g., DeLong or bootstrap) is reported for the difference between HRF external AUC (0.77) and DFR external AUC (~0.57). Without this, the claim that “HRF-based models generalized best” rests on point estimates whose uncertainty overlaps.
minor comments (3)
  1. [Abstract] Abstract: the notation “accuracy 0.77 +/- 0.00” should be clarified; if it reflects a single external partition, state the test-set size explicitly.
  2. [Methods] Methods: list the exact hyperparameter search ranges and final values for all ML models (random forest, etc.), as these are free parameters that affect reproducibility.
  3. [Figures] Figure legends: SHAP summary plots should label the top features by their radiomic name (e.g., “GLCM_Entropy”) rather than generic indices.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and have revised the manuscript to improve transparency, add requested analyses, and strengthen the statistical support for our claims.

read point-by-point responses

  1. Referee: Methods section on semi-supervised pseudo-labeling: the threshold, iteration count, and any quality check against the 136 ground-truth labels are not specified. Because the headline external AUC of 0.77 is obtained only after training on the pseudo-labeled pool, an ablation that reports performance when the model is trained solely on the 136 labeled cases is required to demonstrate that pseudo-labeling does not inject center-correlated label noise.

    Authors: We agree that greater detail on the semi-supervised procedure is needed for reproducibility. In the revised Methods section we now specify the pseudo-labeling threshold, iteration count, and quality-control steps performed against the ground-truth labels. We have also added the requested ablation study comparing models trained only on the 136 labeled cases versus the full pseudo-labeled set; the new results confirm that the semi-supervised step improves external performance without evidence of center-correlated label noise. revision: yes

  2. Referee: Results, external-testing paragraph and associated table: class prevalence in the external test set is not stated, nor is any strategy for three-class imbalance (e.g., class weights, oversampling). The reported accuracy of 0.77 ± 0.00 with zero variance is difficult to interpret without this information and raises the possibility that the metric is dominated by the majority class.

    Authors: We acknowledge that class prevalence and imbalance handling should have been stated explicitly. The revised Results section now reports the class distribution in the external test set and describes the class-weighted loss used to address three-class imbalance. We have also clarified the accuracy metric by reporting its variance across the evaluation folds to facilitate interpretation. revision: yes

  3. Referee: Results, model-comparison subsection: no statistical test (e.g., DeLong or bootstrap) is reported for the difference between HRF external AUC (0.77) and DFR external AUC (~0.57). Without this, the claim that “HRF-based models generalized best” rests on point estimates whose uncertainty overlaps.

    Authors: We agree that a formal statistical comparison is required to support the model ranking. In the revised Results we have added a DeLong test comparing the external AUCs of the HRF and DFR models; the test establishes a statistically significant difference, thereby strengthening the claim that HRF-based models generalized best. revision: yes

Circularity Check

0 steps flagged

No circularity in reported performance metrics or modeling pipeline

full rationale

The paper reports empirical results from five-fold cross-validation and external testing on independent multicenter data (1,023 scans from 12 public datasets). No equations, derivations, or modeling steps are shown that reduce the reported AUC/accuracy to fitted parameters defined on the same test set by construction. The semi-supervised pseudo-labeling is applied only to unlabeled scans, with final metrics evaluated on held-out external data preserving independence. No self-citations, ansatzes, or uniqueness theorems are invoked in a load-bearing way within the provided text. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central performance numbers rest on the domain assumption that CT-derived radiomic phenotypes correlate with underlying genetic mutations and on standard machine-learning fitting procedures whose hyperparameters are not enumerated in the abstract.

free parameters (2)
  • ML model hyperparameters
    Standard classification models require choices for regularization, learning rate, and architecture depth that are fitted during cross-validation.
  • pseudo-labeling threshold
    Semi-supervised strategy uses a threshold to assign labels to unlabeled scans; exact value is not stated.
axioms (1)
  • domain assumption Radiomic features extracted from CT can serve as proxies for EGFR and KRAS mutation status
    Invoked by the entire radiogenomics pipeline and the claim that morphology and heterogeneity features are predictive.

pith-pipeline@v0.9.0 · 5627 in / 1281 out tokens · 42905 ms · 2026-05-15T00:33:17.444980+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages

  1. [1]

    Siegel, T.B.K., Angela N

    Rebecca L. Siegel, T.B.K., Angela N. Giaquinto M, Hyuna Sung, Ahmedin Jemal DVM,, Cancer statistics, 2025. CA Cancer J Clin, 2025. 75(1): p. 10-45

    work page 2025

  2. [2]

    European radiology, 2016

    Rizzo S, P.F., Buscarino V, De Maria F, Raimondi S, Barberis M, et al, CT radiogenomic characterization of EGFR, K - RAS, and ALK mutations in non-small cell lung cancer. European radiology, 2016. 26(1): p. 32-42

    work page 2016

  3. [3]

    Cells, 2023

    Shaban N, K.D., Emelianova A, Buzdin A, Targeted inhibitors of EGFR: structure, biology, biomarkers, and clinical applications. Cells, 2023. 13(1): p. 47

    work page 2023

  4. [4]

    , KRAS mutation is a weak, but valid predictor for poor prognosis and treatment outcomes in NSCLC: A meta-analysis of 41 studies

    Pan W, Y.Y., Zhu H, Zhang Y, Zhou R, Sun X. , KRAS mutation is a weak, but valid predictor for poor prognosis and treatment outcomes in NSCLC: A meta-analysis of 41 studies. Oncotarget, 2016. 7(7): p. 8373

    work page 2016

  5. [5]

    , Literature review of advances and challenges in KRAS G12C mutant non-small cell lung cancer

    Yuan J-X, H.Y., Dai X-Z, Hong J-J, Chen C-Y, Huo Z-X, et al. , Literature review of advances and challenges in KRAS G12C mutant non-small cell lung cancer. Translational Lung Cancer Research, 2025. 14(7): p. 2799

    work page 2025

  6. [6]

    Chaddha U, A.A., Ghori U, Kheir F, Debiane L, McWilliams A, et al. , Safety and Sample Adequacy for Comprehensive Biomarker Testing of Bronchoscopic Biopsies: An American Association of Bronchology and Interventional Pulmonology 13 (AABIP) and International Association for the Study of Lung Cancer (IASLC) Clinical Practice Guideline. Journal of Thoracic O...

    work page 2025

  7. [7]

    Predictive value of 18F -FDG PET/CT radiomics for EGFR mutation status in non-small cell lung cancer: a systematic review and meta-analysis

    Ma N, Y.W., Wang Q, Cui C, Hu Y, Wu Z. . Predictive value of 18F -FDG PET/CT radiomics for EGFR mutation status in non-small cell lung cancer: a systematic review and meta-analysis. Frontiers in Oncology, 2024. 14: p. 1281572

    work page 2024

  8. [8]

    , Construction of a radiogenomics predictive model for KRAS mutation status in patients with non-small cell lung cancer

    Li Y, L.J., Wang Y, Wang Y, Huang D, Wen Z, et al. , Construction of a radiogenomics predictive model for KRAS mutation status in patients with non-small cell lung cancer. . Journal of Thoracic Disease., 2025. 17(6): p. 3749

    work page 2025

  9. [9]

    , Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach

    Aerts HJ, V.E., Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. , Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nature communications, 2014. 5(1): p. 4006

    work page 2014

  10. [10]

    Physics in Medicine & Biology, 2016

    Yip SS, A.H., Applications and limitations of radiomics. Physics in Medicine & Biology, 2016. 61(13): p. R150

    work page 2016

  11. [11]

    -F., Li X -Y, Yu W, Xu Z -Y, Cai X -W, et al

    Jia T-Y, X.J. -F., Li X -Y, Yu W, Xu Z -Y, Cai X -W, et al. , Identifying EGFR mutations in lung adenocarcinoma by noninvasive imaging using radiomics features and random forest modeling. European radiology, 2019. 29(9): p. 4742-50

    work page 2019

  12. [12]

    Molecular Imaging and Biology, 2020

    Shiri S, M.H., Hajianfar Gh, Abdollahi H, Ashrafinia S, Hatt M, Zaidi H, Oveisi M, and Rahmim A, Next -generation radiogenomics sequencing for prediction of EGFR and KRAS mutation status in NSCLC patients using multimodal imaging and machine learning algorithms. Molecular Imaging and Biology, 2020. 22(4): p. 1132-48

    work page 2020

  13. [13]

    Computers in biology and medicine, 2022

    Shiri I, A.M., Nazari M, Hajianfar Gh, Avval A H, Abdollahi H, Oveisi M, Arabi H, Rahmim A, Zaidi H, Impact of feature harmonization on radiogenomics analysis: Prediction of EGFR and KRAS mutations from non -small cell lung cancer PET/CT images. Computers in biology and medicine, 2022. 142: p. 105230

    work page 2022

  14. [14]

    , Somatic mutations drive distinct imaging phenotypes in lung cancer

    Rios Velazquez E, P.C., Liu Y, Coroller TP, Cruz G, Stringfield O, et al. , Somatic mutations drive distinct imaging phenotypes in lung cancer. Cancer research, 2017. 77(14): p. 3922-30

    work page 2017

  15. [15]

    , Predictive radiogenomics modeling of EGFR mutation status in lung cancer

    Gevaert O, E.S., Khuong A, Hoang CD, Shrager JB, Jensen KC, et al. , Predictive radiogenomics modeling of EGFR mutation status in lung cancer. Scientific reports, 2017. 7(1): p. 41674

    work page 2017

  16. [16]

    Korean journal of radiology 2019

    Park J E, P.S.Y., Kim H J, and Kim H S., Reproducibility and generalizability in radiomics modeling: possible strategies in radiologic and statistical perspectives. Korean journal of radiology 2019. 20(7): p. 1124-1137

    work page 2019

  17. [17]

    Radiology: Artificial Intelligence 2024

    Horvat N, P.N., and Koh D-M, Radiomics beyond the hype: a critical evaluation toward oncologic clinical use. Radiology: Artificial Intelligence 2024. 6(4): p. e230437

    work page 2024

  18. [18]

    , A postreconstruction harmonization method for multicenter radiomic studies in PET

    Orlhac F, B.S., Philippe C, Stalla-Bourdillon H, Nioche C, Champion L, et al. , A postreconstruction harmonization method for multicenter radiomic studies in PET. Journal of Nuclear Medicine, 2018. 59(8): p. 1321-8

    work page 2018

  19. [19]

    , The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping

    Zwanenburg A, V.M., Abdalah MA, Aerts HJ, Andrearczyk V, Apte A, et al. , The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology, 2020. 295(2): p. 328-38

    work page 2020

  20. [20]

    Felfli, M., Liu Y, Zerka F, Voyton Ch, Thinnes A, Jacques S, Iannessi A, et al. , Systematic review, meta -analysis and radiomics quality score assessment of CT radiomics-based models predicting tumor EGFR mutation status in patients with non-small-cell lung cancer. International journal of molecular sciences, 2023. 24(14): p. 11433

    work page 2023

  21. [21]

    , Multi -channel multi-task deep learning for predicting EGFR and KRAS mutations of non-small cell lung cancer on CT images

    Dong Y, H.L., Yang W, Han J, Wang J, Qiang Y, Zhao J et al. , Multi -channel multi-task deep learning for predicting EGFR and KRAS mutations of non-small cell lung cancer on CT images. Quantitative imaging in medicine and surgery,

    work page

  22. [22]

    Journal of Cancer Research and Clinical Oncology

    Zuo Y, L.Q., Li N, Li P, Fang Y, Bian L, Zhang J, Song S, Explainable 18F-FDG PET/CT radiomics model for predicting EGFR mutation status in lung adenocarcinoma: a two -center study. Journal of Cancer Research and Clinical Oncology

    work page

  23. [23]

    K, N.N., Kennedy E B, Biermann W A, Donington J, Leighl N B, Lew M et al

    Gregory P. K, N.N., Kennedy E B, Biermann W A, Donington J, Leighl N B, Lew M et al. , Molecular testing guideline for the selection of patients with lung cancer for treatment with targeted tyrosine kinase inhibitors: American Society of Clinical Oncology Endorsement of the College of American Pathologists/International Association for the Study of Lung C...

    work page 2018

  24. [24]

    , Actionable mutations and targeted therapy in non -small cell lung cancer among Latin American and Hispanic patients: a systematic literature review of prognosis and meta-analysis

    González C, P.J., Riess J W, Gómez-Gómez M P,Clavijo Cabezas D,Vargas M P et al. , Actionable mutations and targeted therapy in non -small cell lung cancer among Latin American and Hispanic patients: a systematic literature review of prognosis and meta-analysis. Translational Lung Cancer Research, 2025. 14(9): p. 3410-3429

    work page 2025

  25. [25]

    Diagnostic and Interventional Radiology

    D, A., Reproducibility and interpretability in radiomics: a critical assessment. Diagnostic and Interventional Radiology

    work page

  26. [26]

    European Radiology Experimental, 2024

    Leonardo R, M.C., Image biomarkers and explainable AI: handcrafted features versus deep learned features. European Radiology Experimental, 2024. 8(1): p. 130

    work page 2024

  27. [27]

    The cancer genome atlas lung adenocarcinoma collection (tcga-luad)

    Albertina B, W.M., Holback C, Jarosz R, Kirk S, Lee Y, et al. The cancer genome atlas lung adenocarcinoma collection (tcga-luad). (No Title). 2016., The cancer genome atlas lung adenocarcinoma collection (tcga-luad). 2016

    work page 2016

  28. [28]

    , Data for NSCLC Radiogenomics collection

    Bakr S, G.O., Echegaray S, Ayers K, Zhou M, Shafiq M, et al. , Data for NSCLC Radiogenomics collection. . The Cancer Imaging Archive, 2017

    work page 2017

  29. [29]

    , SPIE -AAPM- NCI Lung Nodule Classification Challenge Dataset, in The Cancer Imaging Archive

    Armato III, S.L., Tourassi G D, Drukker K, Giger M L, Li, Feng R G, Farahani K, Lirby J S, Clarke LP. , SPIE -AAPM- NCI Lung Nodule Classification Challenge Dataset, in The Cancer Imaging Archive. [Available from: https://doi.org/10.7937/K9/TCIA.2015.UZLSU3FL, T.C.I. Archive., Editor. 2015

    work page doi:10.7937/k9/tcia.2015.uzlsu3fl 2015

  30. [30]

    Zhao B, S.L.H., Kris M G, Riely G J, Coffee-break lung ct collection with scan images reconstructed at multiple imaging parameters. 2015

    work page 2015

  31. [31]

    The Cancer Imaging Archive (TCIA)

    Pilot, R., Lung Image Database Consortium (LIDC). The Cancer Imaging Archive (TCIA). 2023

    work page 2023

  32. [32]

    , Data from QIN lung CT

    Goldgof D, H.L., Hawkins S, Schabath M, Stringfield O, Garcia A, et al. , Data from QIN lung CT. 2015

    work page 2015

  33. [33]

    Quantitative computed tomographic descriptors associate tumor shape complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma

    Grove O, B.A., Schabath MB, Aerts HJ, Dekker A, Wang H, et al. Quantitative computed tomographic descriptors associate tumor shape complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma. PloS one. 2015;10(3):e0118261., Quantitative computed tomographic descriptors associate tumor shape complexity and intratumor heterogeneity with pr...

    work page 2015

  34. [34]

    , The lung image database consortium (LIDC) and image database resource initiative (IDRI): a completed reference database of lung nodules on CT scans

    Armato III, S.L., McLennan G, Bidaut L, McNitt‐Gray MF, Meyer CR, Reeves AP, et al. , The lung image database consortium (LIDC) and image database resource initiative (IDRI): a completed reference database of lung nodules on CT scans. . Medical physics, 2011. 38(2): p. 915-31

    work page 2011

  35. [35]

    The Cancer Imaging Archive, 2018

    Madabhushi A, R.M., Fused radiology-pathology lung dataset. The Cancer Imaging Archive, 2018

    work page 2018

  36. [36]

    , Data from lung CT segmentation challenge

    Yang J, S.G., Veeraraghavan H, Van Elmpt W, Dekker A, Lustberg T, et al. , Data from lung CT segmentation challenge. . The cancer imaging archive, 2017

    work page 2017

  37. [37]

    Albertina, M.W., V

    B. Albertina, M.W., V. Holback, R. Jarosz, S. Kirk, Y. Lee, K. Rieger-Christ and J. Lemmerman, , "The Cancer Genome Atlas Lung Adenocarcinoma Collection (TCGA-LUAD) (Version 4) [Data set]. 2016

    work page 2016

  38. [38]

    , PySERA: Open - Source Standardized Python Library for Automated, Scalable, and Reproducible Handcrafted and Deep Radiomics

    Salmanpour, M.R., Pouria A H, Barichin S, Salehi Y, Falahati S, Shiri I, Oveisi M, and Rahmim A. , PySERA: Open - Source Standardized Python Library for Automated, Scalable, and Reproducible Handcrafted and Deep Radiomics. arXiv preprint arXiv:2511.15963, 2025

    work page arXiv 2025

  39. [39]

    EJNMMI physics, 2025

    Du Dongyang, S.I., Yousefirizi F, Salmanpour M R, Lv J, Wu H, Zhu W, Zaidi H, Lu L, Rahmim A Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET- based prediction of lung cancer subtypes. EJNMMI physics, 2025. 12(1): p. 34

    work page 2025

  40. [40]

    , Machine learning evaluation metric discrepancies across programming languages and their components in medical imaging domains: need for standardization

    Salmanpour M R, A.M., Ghazal Mousavi Gh, Sadeghi S, Amiri S, Oveisi M, Rahmim A, d Hacihaliloglu I. , Machine learning evaluation metric discrepancies across programming languages and their components in medical imaging domains: need for standardization. IEEE Access 2025

    work page 2025

  41. [41]

    Oveisi, S

    Alizadeh, M., M. Oveisi, S. Falahati, G. Mousavi, M. A. Meybodi, S. S. Mehrnia, I. Hacihaliloglu, A. Rahmim, and M. R. Salmanpour. , AllMetrics: A Unified Python Library for Standardized Metric Evaluation in Machine Learning. In 2025 IEEE Nuclear Science Symposium (NSS), Medical Imaging Conference (MIC) and Room Temperature Semiconductor Detector Conferen...

    work page 2025

  42. [42]

    , Enhancement Without Contrast: Stability -Aware Multicenter Machine Learning for Glioma MRI Imaging

    Amiri S, T.S., Gharibi S, Dehghanfard S, Mehrnia S S, Oveisi M, Hacihaliloglu I, Rahmim A, Salmanpour MR. , Enhancement Without Contrast: Stability -Aware Multicenter Machine Learning for Glioma MRI Imaging. Inventions

    work page

  43. [43]

    Machine Learning and Knowledge Extraction 2024

    Ribeiro SM, G.A., Sanchez -Gendriz I, SHapley additive explanations (SHAP) for efficient feature selection in rolling bearing fault diagnosis. Machine Learning and Knowledge Extraction 2024. 6(1): p. 316-341

    work page 2024

  44. [44]

    2025:, Radiomics based on MRI and 18F -FDG PET/CT predicts response to EGFR -TKI therapy based on primary NSCLC and brain metastasis

    Fan Y, Y.C., Hu Y, Zhao P, Sun Y, Jiang M, et al. . 2025:, Radiomics based on MRI and 18F -FDG PET/CT predicts response to EGFR -TKI therapy based on primary NSCLC and brain metastasis. Neuro -Oncology Advances, 2025: p. vdaf100

    work page 2025

  45. [45]

    , Radiomic detection of EGFR mutations in NSCLC

    Rossi G, B.E., Fedeli A, Ficarra G, Coco S, Russo A, et al. , Radiomic detection of EGFR mutations in NSCLC. Cancer Research, 2021. 81(3): p. 724-731

    work page 2021

  46. [46]

    , Radiomics signature: a potential and incremental predictor for EGFR mutation status in NSCLC patients, comparison with CT morphology

    Tu W, S.G., Fan L, Wang Y, Xia Y, Guan Y, et al. , Radiomics signature: a potential and incremental predictor for EGFR mutation status in NSCLC patients, comparison with CT morphology. Lung Cancer, 2019. 132: p. 28-35

    work page 2019

  47. [47]

    Radiotherapy and Oncology, 2024

    Chen J, C.A., Yang S, Liu J, Xie C, Jiang H, Accuracy of machine learning in preoperative identification of genetic mutation status in lung cancer: A systematic review and meta-analysis. Radiotherapy and Oncology, 2024. 196: p. 110325

    work page 2024

  48. [48]

    European Radiology Experimental, 2023

    A, D., Are deep models in radiomics performing better than generic models? A systematic review. European Radiology Experimental, 2023. 7(1): p. 11

    work page 2023

  49. [49]

    Insights into Imaging, 2022

    A, D., Predictive performance of radiomic models based on features extracted from pretrained deep networks. Insights into Imaging, 2022. 13(1): p. 187

    work page 2022

  50. [50]

    Translational Cancer Research, 2018

    Vial A, S.D., Field M, Ros M, Ritz C, Carolan M, et al., The role of deep learning and radiomic feature extraction in cancer-specific predictive modelling: a review. Translational Cancer Research, 2018. 7(3)

    work page 2018

  51. [51]

    Emerging topics in life sciences, 2021

    P, W., Radiomics, deep learning and early diagnosis in oncology. Emerging topics in life sciences, 2021. 5(6): p. 829-35

    work page 2021

  52. [52]

    , Hybrid Approach to Classifying Histological Subtypes of Non-small Cell Lung Cancer (NSCLC): Combining Radiomics and Deep Learning Features from CT Images

    Oh G, G.Y., Lee J, Kim H, Wu H-G, Park JM, et al. , Hybrid Approach to Classifying Histological Subtypes of Non-small Cell Lung Cancer (NSCLC): Combining Radiomics and Deep Learning Features from CT Images. Journal of Imaging Informatics in Medicine, 2025: p. 1-13

    work page 2025

  53. [53]

    , Deep learning –radiomics integrated noninvasive detection of epidermal growth factor receptor mutations in non-small cell lung cancer patients

    Kim S, L.J., Kim C -H, Roh J, You S, Choi J -S, et al. , Deep learning –radiomics integrated noninvasive detection of epidermal growth factor receptor mutations in non-small cell lung cancer patients. Scientific Reports, 2024. 14(1): p. 922

    work page 2024

  54. [54]

    Diagnostics 2023

    A, D., Deep features from pretrained networks do not outperform hand -crafted features in radiomics. Diagnostics 2023. 13(20): p. 3266

    work page 2023