Intelligent medical diagnosis and treatment for diabetes with deep convolutional fuzzy neural networks

The advent of smart healthcare has significantly heightened the importance of computer technologies in supporting medical diagnosis and treatment. Nevertheless, the challenges of mining latent knowledge within diagnostic data and explaining results to healthcare professionals have limited the application of many algorithms in clinical practice. To address these issues, our study introduces an Interpretable Predictor with Deep Convolutional Fuzzy Neural Network (IP-DCFNN). The proposed model is capable of assessing disease risk based on individual data and providing interpretable justifications to aid in medical diagnosis and treatment decisions. By deconstructing the fuzzy inference process and incorporating convolutional neural network, our approach enhances the ability to discover underlying information while maintaining transparency and interpretability. Furthermore, we introduce a grid partition-based method for initializing the antecedent parameters and a hybrid approach that combines gradient descent with least squares estimation for training. Compared with Adaptive Neuro-Fuzzy Inference System (ANFIS) and Deep Neural Networks (DNN) Our model has an average improvement of 7.4% on prediction accuracy. More importantly, it can extract interpretable insights from membership functions, rule bases, and fuzzy contributions, offering valuable knowledge for medical research on type 2 diabetes, supporting intelligent diagnostic processes and providing personalized healthcare recommendations. The model can also be applied on the diagnosis and treatment of various other diseases.