Explainable Diagnosis Prediction through Neuro-Symbolic Integration

TL;DR

The paper tackles the need for explainable diagnosis prediction in healthcare by integrating domain knowledge with learning through neuro-symbolic methods. It introduces Logical Neural Networks (LNNs) that embed differentiable logic operators and learnable thresholds, producing interpretable rules for diagnosis. The strongest LNN-based models, $M_{multi-pathway}$ and $M_{comprehensive}$, outperform traditional baselines (e.g., Logistic Regression, SVM, Random Forest) with accuracies up to $80.52\%$ and AUROCs up to $0.8457$, while providing transparent, weight-based feature contributions. This work demonstrates a viable path toward precision medicine, combining predictive power with interpretability and potential adaptability to diverse populations and conditions.

Abstract

Diagnosis prediction is a critical task in healthcare, where timely and accurate identification of medical conditions can significantly impact patient outcomes. Traditional machine learning and deep learning models have achieved notable success in this domain but often lack interpretability which is a crucial requirement in clinical settings. In this study, we explore the use of neuro-symbolic methods, specifically Logical Neural Networks (LNNs), to develop explainable models for diagnosis prediction. Essentially, we design and implement LNN-based models that integrate domain-specific knowledge through logical rules with learnable thresholds. Our models, particularly $M_{\text{multi-pathway}}$ and $M_{\text{comprehensive}}$, demonstrate superior performance over traditional models such as Logistic Regression, SVM, and Random Forest, achieving higher accuracy (up to 80.52\%) and AUROC scores (up to 0.8457) in the case study of diabetes prediction. The learned weights and thresholds within the LNN models provide direct insights into feature contributions, enhancing interpretability without compromising predictive power. These findings highlight the potential of neuro-symbolic approaches in bridging the gap between accuracy and explainability in healthcare AI applications. By offering transparent and adaptable diagnostic models, our work contributes to the advancement of precision medicine and supports the development of equitable healthcare solutions. Future research will focus on extending these methods to larger and more diverse datasets to further validate their applicability across different medical conditions and populations.

Figures (1)

• Figure 1: Visualization of feature contributions in the rule models $M_{\text{multi-pathway}}$ and $M_{\text{comprehensive}}$ for diabetes prediction. The features are denoted as follows: G (glucose), I (insulin), B (BMI), S (skin thickness), T (blood pressure), D (DPF), and A (age). Both diagrams represent the logical structures of the models, where the leaf nodes correspond to features with their respective learned thresholds displayed below each node. The edges connecting the nodes are labeled with weights, indicating the relative importance of each feature in the prediction process. The combination of feature weights and thresholds demonstrates how different pathways contribute to the final prediction in each model.