Interpretable Machine Learning with Brain Image and Survival Data
چکیده
Recent developments in research on artificial intelligence (AI) medicine deal with the analysis of image data such as Magnetic Resonance Imaging (MRI) scans to support decision-making medical personnel. For this purpose, machine learning (ML) algorithms are often used, which do not explain internal process at all. Thus, it is difficult validate or interpret results applied AI methods. This manuscript aims overcome problem by using methods explainable (XAI) an ML algorithm use case predicting survival rate patients brain tumors based MRI scans. Therefore, we explore images together predict gliomas a focus improving interpretability results. Using Brain Tumor Segmentation dataset BraTS 2020, used well-validated for evaluation and relied convolutional neural network structure improve explainability important features adding Shapley overlays. The trained models were evaluate SHapley Additive exPlanations (SHAP) directly optimized accuracy. resulting overfitting some structures therefore seen presented interpretation method. It shown that can be validated experts visualizations, thus making method interpretable. Our study highlights feasibility combining explainers 3D voxels also fact prediction significantly supports implementation python available gitlab “XAIforBrainImgSurv”.