Toward Neuro-Symbolic and Reservoir-Inspired Medical Imaging: A BD-CeNN Autoencoder with ASP Rule Mining for Robust and Explainable Interpretation of Grayscale and Color Images

Abstract

We present a neuro-symbolic framework for medical image analysis that integrates a Binary Discrete Cellular Neural Network (BD-CeNN) autoencoder, a reservoir-computing–inspired BD-CeNN refinement stack, and automatically mined Answer Set Programming (ASP) rules. The autoencoder converts grayscale and color inputs (CT, MRI, histopathology, dermatology) into discrete, symbolic latent codes, which are iteratively refined to improve robustness and diagnostic discrimination. From annotated cases, ASP rules capture human-readable relations and constraints, enabling transparent, auditable reasoning over the learned symbols while maintaining predictive performance. The hybrid design targets resource-constrained clinical environments where trust, explainability, and adaptability are essential. This paper details the conceptual architecture, motivation, and deployment feasibility; extensive benchmarking is left for future work, laying the groundwork for accessible, interpretable AI in medical imaging.