Development of a multimodal fully automated ensemble model to predict EGFR mutation and efficacy of EGFR-TKI in non-small cell lung cancer

Sep 29, 2026

Uploaded by Asher

Study investigates an artificial intelligence approach to detect epidermal growth factor receptor (EGFR) mutation and predict treatment efficacy of EGFR-tyrosine kinase inhibitors (EGFR-TKIs) in non-small cell lung cancer (NSCLC), where conventional testing relies on invasive, time-consuming tumor biopsies and therapeutic benefits vary widely. An ensemble model integrates machine learning for clinical information and deep convolutional neural networks for computed tomography (CT) images, trained to predict EGFR mutation and response using cross-entropy loss with 5-fold cross-validation. Results show strong mutation prediction performance (AUC 0.88) and better treatment-response prediction for exon 19 deletions than L858R mutations, supporting non-invasive, genotype-specific outcome prediction.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Highlight box

Key findings

What is known and what is new?

Methods

Results

Conclusions

Development of a multimodal fully automated ensemble model to predict EGFR mutation and efficacy of EGFR-TKI in non-small cell lung cancer Shiting Xu1,2\#, Taichi Miyawaki2\#, Takehito Shukuya2, Kazuhiro Suzuki3, Shoko Sonobe Shimamura2, Hironari Matsuda2, Ryota Kanemaru2, Tetsuhiko Asao2, Tomoyasu Mimori2, Yujiro Otsuka3,4,5, Kazuhisa Takahashi2 1Endoscopy Center, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai, China; 2Department of Respiratory Medicine, Juntendo University Graduate School of Medicine, Tokyo, Japan; 3Department of Radiology, Juntendo University Faculty of Medicine and Graduate School of Medicine, Tokyo, Japan; 4Plusman LLC., Tokyo, Japan; 5Milliman, Inc., Tokyo, Japan Contributions: (I) Conception and design: T Shukuya; (II) Administrative support: K Takahashi; (III) Provision of study materials or patients: TMiyawaki, K Suziki, SS Shimamura, H Matsuda, R Kanemaru, T Asao, T Mimori; (IV) Collection and assembly of data: S Xu, T Miyawaki, T Shukuya; (V) Data analysis and interpretation: Y Otsuka, T Shukuya; (VI) Manuscript writing: All authors; (VII) Final approval of manuscript: All authors. \#These authors contributed equally to this work. Correspondence to: Takehito Shukuya, MD, PhD. Department of Respiratory Medicine, Juntendo University Graduate School of Medicine, 3-1-3 Hongo, Bunkyo-Ku, Tokyo 113-8421, Japan. Email: [tshukuya@juntendo.ac.jp](tshukuya@juntendo.ac.jp). Background: Drugs for systemic therapy to advanced or recurrent non-small cell lung cancer (NSCLC) are determined by the type of driver oncogenes and programmed death-ligand 1 (PD-L1) immunostaining. In this study, we investigated the potential for artificial intelligence to assist in detecting epidermal growth factor receptor (EGFR) mutation, which is one of the most frequent and important driver oncogenes, and predicting the efficacy of EGFR-tyrosine kinase inhibitors (TKIs) in patients with NSCLC. Methods: We built an ensemble model that combined machine learning models (logistic regression, Bernoulli naive Bayes and Gaussian naive Bayes) for clinical information and tumor characteristics analysis, and a deep convolutional neural network for analyzing computed tomography (CT) images. The convolutional neural network was trained from scratch on our imaging dataset. Clinical and imaging features were analyzed separately within their respective models and subsequently integrated in the ensemble framework. The models were trained to predict EGFR mutation and the response of EGFR-TKI using cross-entropy loss. The performance of the proposed model was evaluated by 5-fold cross-validation. Results: One hundred and fifty consecutive evaluable patients were included in this study. Among them, 61 patients had EGFR mutation, including 59 patients with common EGFR mutation and 2 patients with uncommon EGFR mutation. The ensemble model predicted EGFR mutation with an area under the curve (AUC) of 0.88 [95% confidence interval (CI): 0.79–0.97] . Each sub-model integrated into the ensemble model predicted EGFR mutation with AUC of 0.83 (95% CI: 0.71–0.95), 0.88 (95% CI: 0.79–0.97), 0.87 (95% CI: 0.79–0.94), and 0.84 (95% CI: 0.72–0.96) for logistic regression, Bernoulli naive Bayes, Gaussian naive Bayes and deep neural network respectively. The predictive accuracy of the ensemble model was 0.56. Specifically, the ensemble model reported that the predictive accuracy of the Del19 EGFR mutation was higher than that ofthe L858R EGFR mutation, with overall response rate predictive accuracy (AUC) of 0.67 versus 0 . 52, and disease control rate AUC of 0.88 versus 0.53. Conclusions: The ensemble model demonstrated strong performance in predicting EGFR mutations and showed higher predictive treatment response for EGFR exon 19 deletions than for L858R mutations. These findings suggest that ensemble learning may enhance the non-invasive prediction of EGFR mutation status and EGFR genotype-specific the

FAQs

What problem does the study address in NSCLC treatment?

EGFR mutation testing typically requires invasive tumor biopsies, and patient responses to EGFR-TKIs vary. The study targets non-invasive prediction of both EGFR mutation status and EGFR-TKI efficacy.

How is the proposed AI model structured?

The model is an ensemble combining machine learning for clinical information and a deep convolutional neural network for computed tomography (CT) images. Clinical and imaging features are analyzed separately and integrated in the ensemble framework.

What were the main performance and outcome findings?

The ensemble model predicted EGFR mutation with an AUC of 0.88 and showed higher predictive accuracy for therapeutic response in EGFR exon 19 deletions than in L858R mutations.

Sep 29, 2026

Uploaded by Asher