Paper Details

Name: Bhawana Goyal, Mukhtiar Singh & Munish Kumar
Country: India

Year: 2026
Volume: Volume No: 13, January, Year: 2026 (Special Issue)
Page Number: 779-786
DOI: https://doi.org/10.5281/zenodo.19129667

ABSTRACT
Medical conditions are generally classified into general and rare medical conditions. Rare medical diseases such as Huntington’s Disease, Idiopathic Pulmonary Fibrosis, Mesothelioma often are neglected in the field of automotive prediction due to the data scarcity. In this paper a novel neuro- symbolic approach is designed to synthesize clinically plausible patient records by mixing deep learning along with domain specific symbolic reasoning. Approach involves building a framework upon Variational Autoencoder (VAE) augmented with symbolic constraint modules that enforces
strict clinical guidelines which ensures that the generated synthetic data adhere to the established medical norms. A detailed and comprehensive approach is encompassed by fixating on multiple parameters and then integrating the symbolic constraints for loss functions. Extensive experiments on longitudinal clinical data suggests that the proposed neurosymbolic generative model not only mitigates the limitations posed by rare disease datasets but also enhances downstream task by providing with high quality of synthetic data. A robust and safe interpretable framework is developed paving the way for improved diagnostic and prognostic tools in rare medical disease conditions.

General Terms : Data Synthesis, concordance, Artificial Intelligence, Healthcare, Generative AI Learning.

Keywords: VAE, Encoder, NS-VAE, Deep Learning, Generative AI.