Non-Small Cell Lung Cancer Coverage from Every Angle
Advertisement
Advertisement

EGFR-TKI Therapy for Lung Cancer: Can Machine Learning Model Identify Patients Unlikely to Benefit?

By: Anna Nowogrodzki
Posted: Tuesday, March 30, 2021

A deep learning semantic signature seemed to be better than radiomics signatures at both predicting progression-free survival in patients with stage IV EGFR variant–positive non–small cell lung cancer (NSCLC) and identifying those unlikely to benefit from EGFR–tyrosine kinase inhibitor (TKI) therapy. The recent diagnostic/prognostic study was published by Jie Tian, PhD, of Beihang University, Beijing, China, and colleagues in JAMA Network Open.

“The end-to-end deep learning–derived semantic features eliminated all manual interventions required while using previous radiomics methods and have a better prognostic performance,” the authors wrote.

The prognostic study included 342 patients from 2010 to 2017. Patients were all receiving EGFR-TKI therapy for stage IV EGFR variant–positive NSCLC. Using a training cohort of 145 of the patients, the authors built a deep learning semantic signature to predict progression-free survival. They validated the deep learning model by having it predict progression-free survival in two external validation cohorts (197 patients total); they then compared these results with survival predictions based on radiomics signatures. The deep learning model used patients’ original whole CT scan images as input.

The deep learning semantic signature identified 90 patients (26%) who were at high risk of rapid disease progression on EGFR-TKI therapy. Clinical decisions based on the deep learning semantic signature’s predictions of tumor progression risk resulted in better survival outcomes than decisions based on radiomics signatures. This result held true across all risk probabilities, the study authors noted.

The use of this deep learning approach eliminated the labor-intensive manual steps needed to use radiomics signatures, such as manually segmenting tumor boundaries. The source code is open-access, so anyone can use it to replicate the results of the study, according to the investigators.

Disclosure: The study authors reported no conflicts of interest.



By continuing to browse this site you permit us and our partners to place identification cookies on your browser and agree to our use of cookies to identify you for marketing. Read our Privacy Policy to learn more.