Using MACHINE LEARNING to Predict Response to Pemetrexed Chemotherapy

Overview

USING MACHINE LEARNING AND RADIOMICS TEXTURE FEATURES TO PREDICT RESPONSE OF ADENOCARCINOMA PATIENTS TO PEMETREXED CHEMOTHERAPY Jessica Chang, [email protected], Pranjal Vaidya, [email protected], Kaustav Bera, [email protected] Anant Madabhushi, [email protected], 1Hathaway Brown School, Shaker Heights OH 44122, 2Case Western University, Dept of Biomedical Engineering, Cleveland OH 44106 Adenocarcinoma is one of the most common subtypes of non-small cell lung cancer, or NSCLC, with an average five-year survival rate of around 18%. A key component of chemotherapy treatment for advanced adenocarcinomas is pemetrexed, an antifolate agent. However, previous studies of the efficacy of pemetrexed in NSCLC have shown that the response rate to pemetrexed therapy is as low as 9.1% (Felip and Rosell). Pemetrexed therapy is expensive and can have detrimental side effects, including nausea, vomiting, fatigue, and stomatitis. This experiment used computer extracted texture features to predict patient response to pemetrexed chemotherapy. A total of N=95 cases were used, with N=35 responsive and N=60 non-responsive to treatment. A total of 5994 texture features were computed for each patient. The top features were then selected using the Wilcoxon test. Two supervised classifiers, LDA (linear discriminant analysis) and QDA (quadratic discriminant analysis) were trained to distinguish between responsive and non-responsive patients. Three-fold classification was then performed using the top 4 features and with 100 iterations. The LDA classifier yielded a mean area under the curve (AUC) of 0.7721, while the QDA classifier yielded a mean AUC of 0.7773. These results suggest supervised radiomics-based machine learning models can be used to predict patient response rates with a high level of precision and accuracy.

Competition history

• AJAS 2020 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• AJAS None projects
• Every AJAS 2020 project