Explainable deep learning framework for classifying pathogen-induced chest infections with fuzzy severity reasoning

Abstract

Accurate classification of pathogen-induced chest infections is a vital step in enabling rapid diagnosis in the context of a high clinical burden scenario. This paper proposes an explainable deep learning (DL) framework that employs a hybrid approach by integrating an efficient classification model with both decision-level and feature-level interpretability. A neural search architecture is adopted to develop a fine-tuned EfficientNet model for multi-class classification of chest radiographs subjected to seven classes. To enhance clinical trust, a Mamdani-type fuzzy-logic-based severity reasoning layer is introduced to provide contextual assessments of disease severity and possible co-infections. Furthermore, feature-level explanations are enabled using gradient-weighted class activation mapping (Grad-CAM) to visualize important regions that contribute to each prediction. The model is validated on a comprehensive dataset, and the experimental outcome demonstrated the effectiveness of the proposed framework with classification accuracy of 96% and sensitivity of 95%.