Attentive Hierarchical ANFIS with interpretability for cancer diagnostic

Abstract

Prediction of clinical outcomes using the patient’s medical data enhances clinical decision making and improves prognostic accuracy. Deep learning (DL) for medical decision support systems has particularly shown expert-level accuracy in predicting clinical outcomes. However, most of these machine learning and Artificial Intelligent (AI) models lack interpretability which causes major trust-related problems in healthcare. This necessitates the need for interpretable AI systems that can explain their decisions. This paper is designed to address the problem of how clinical decision support systems can be designed to be transparent, interpretable and therefore comprehensible for humans. In this view, this paper proposes an attentive hierarchical adaptive neuro-fuzzy inference system (AH-ANFIS), that combines fuzzy inference in a hierarchical architecture with attention for selecting the important rules. The proposed model benefits from the rule-based structure of ANFIS that enables the user to interpret the abstractions of hidden layers. The hierarchical structure in fuzzy modeling helps to overcome the rule explosion problem that arises with large medical data and improve interpretability. To ensure the improvement of interpretability with hierarchical architecture, an evolutionary algorithm (EA) is used to decompose the input space into an optimal permutation of input subsets which results in subsystems with independent meaning. The attention-based rule selector identifies the most activated rule to select important patient-specific features for predicting the clinical outcome. To verify the performance of the proposed AH-ANFIS and analyze the important features for classification, we perform experiments with two cancer diagnostic datasets. By pruning fuzzy if-then rules using recursive rule elimination (RRE), the complexity of the model is largely reduced while maintaining the performance of the system making it more interpretable.