An interpretable machine learning model assists in predicting induction chemotherapy response and survival for
Hai Liao1, Yang Zhao1, Wei Pei1
1Department of Medical Imaging Center, Guangxi Medical University Cancer Hospital, Guangxi Clinical Research Center for Medical Imaging Construction, Guangxi Key Clinical Specialty (Medical Imaging Department), Dominant Cultivation Discipline of Guangxi Medical University Cancer Hospital (Medical Imaging Department), Nanning, China.
Objectives:
To develop and validate an interpretable and generalized machine learning model using MRI for the individualized prediction of induction chemotherapy (ICT) response and survival in locoregionally advanced nasopharyngeal carcinoma (LANPC).
Methods:
A total of 1368 patients who underwent MRI examinations before ICT from three hospitals were retrospectively enrolled and divided into training, internal validation, external validation, and cross-field strength validation cohorts. Significant radiomics and clinical features were selected from coarse to fine. An interpretable genetic algorithm-enhanced artificial neural network (GNN) was applied for models' development and validation. The performance of junior and senior doctors in predicting ICT response with and without model aid was evaluated.
Results:
The interpretable GNN model achieved good generalization performance in predicting ICT response, with areas under the curve (AUCs) ranging from 0.808 to 0.864 across all cohorts. Survival analysis demonstrated that low-risk patients defined by GNN-radiomics signature and clinical factors had better progression-free survival than high-risk patients in all cohorts (hazard ratio ranging from 3.231 to 12.787, p < 0.05). The predictive performance of junior and senior doctors for ICT response significantly improved with model assistance (AUCs: 0.686 vs. 0.785 and 0.736 vs. 0.836, p < 0.05).
Conclusion:
An interpretable, applicable, and generalized GNN model based on multi-center databases achieved superior performance in predicting ICT response and survival in LANPC patients, which may contribute to the personalized treatment of LANPC.
Key Points:
Question Currently, there is a lack of accurate methods for predicting and evaluating the efficacy and prognosis of nasopharyngeal carcinoma (NPC). Findings Genetic algorithm-enhanced artificial neural network model excels in predicting induction chemotherapy response and survival outcome of NPC, providing valuable assistance to doctors in clinical practice. Clinical relevance This model can identify patients likely to benefit from induction chemotherapy, promoting individualized treatment and optimizing clinical management.
More Related Videos
07:15Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
09:53Quantifying the Brain Metastatic Tumor Micro-Environment using an Organ-On-A Chip 3D Model, Machine Learning, and Confocal Tomography
Published on: August 16, 2020