An Explainable Hybrid ConvNeXt-Temporal Fusion Transformer Framework for Semantic Patient Information Retrieval in Distributed Healthcare Systems

The massive growth of Electronic Health Records (EHRs) and clinical healthcare databases poses several problems in the effective and accurate retrieval of patient information. Traditional patient information retrieval mechanisms have found it difficult to incorporate the intricate semantic relationships and temporal dependencies involved in the heterogeneous data to retrieve accurate patient information. With this in mind, we propose an Explainable Hybrid ConvNeXt-Temporal Fusion Transformer Framework for Semantic Patient Information Retrieval (XHCNXT-TFT-PIR) in distributed healthcare systems. First, the healthcare data will be preprocessed by performing data cleaning, imputation, normalization, and feature selection. Second, the ConvNeXt Network will be applied in the extraction of the hierarchical and discriminatory features from the patient information. Third, the extracted features will be processed using Temporal Fusion Transformer (TFT) to model the long-range temporal dependencies of the longitudinal data. To enhance explainability with regard to interpretability, SHAP-based Explainable Artificial Intelligence (XAI) will be employed to determine the key clinical features that play an important role in the patient information retrieval process. Moreover, Artificial Rabbits Optimization (ARO) algorithm will be used in the optimization of the hyperparameters of the models. Experimental evaluation is conducted using the MIMIC-III Clinical Database, containing more than 40,000 patient records and over 50 clinical attributes. The proposed XHCNXT-TFT-PIR framework achieves superior performance with precision 98.92%, recall 98.15%, F1-Score 98.53%, Retrieval Accuracy 99.08%, and Retrieval Latency 72 ms, outperforming existing deep learning and retrieval-based approaches. The results demonstrate the effectiveness of integrating ConvNeXt, Temporal Fusion Transformer, Explainable AI, and Artificial Rabbits Optimization for intelligent and semantically aware patient information retrieval in healthcare big data environments.