ClinNetDx: a novel explainable ai framework for acute lymphoblastic leukemia detection using multi-modal clinical data.

Early identification of Acute Lymphoblastic Leukemia (ALL) is vital to respond on time; yet, presently existing diagnostic procedures still depend on laboratory knowledge, manual interpretation and many supplementary exams. This paper introduces ClinNetDx, a new hybrid deep learning model for Acute Lymphoblastic Leukaemia prediction that combines a Convolutional Neural Network (CNN) and a Deep Neural Network (DNN) and employs structured blood test data and (optional) medical images to support good health and well-being. The proposed methodology includes a tabular-to-spatial transformation that converts numerical clinical measures into 2D representations, enabling a CNN to learn spatial correlations among features from structured medical data. We use advanced pre-processing methods (normalization, imputation, SMOTE) that are robust to class imbalance and missing variables. For interpretability, we use dual explainability: SHAP values to explain feature contributions for numerical data and Grad-CAM visualizations to generate localized heatmaps for image inputs. To make it accessible in a clinical environment, the system is built as a Django-powered web application that provides on-the-fly predictions and an intuitive presentation of findings. The design is flexible and deployed as a RESTful API for interface with medical information systems, mHealth apps and cloud platforms. To demonstrate the promising predictive and generalizing power of the proposed system, it was evaluated on publicly available multi-modal haematological datasets comprising 5,000 structured CBC records, 2,500 ALL smear images, and 2,000 synchronized multi-modal samples using k-fold cross-validation, with confusion matrices and ROC-AUC reported. Experimental assessment shows that ClinNetDx achieves 98.1% classification accuracy, outperforming conventional machine learning and unimodal deep learning models while maintaining strong precision, recall, and ROC-AUC performance. The proposed system combines accurate multi-modal prediction with interpretable decision support, making it a potential option for computer-assisted early diagnosis of Acute Lymphoblastic Leukaemia and for future clinical decision-support applications.