MedScope AI: A Multimodal Deep Learning Framework for Medical Diagnosis and Risk Prediction

Amarilys González García, R. Regin, E. Delhi Saminathan, S. Nadeemul Haq, R. Darshan, A. T. Ashmi Christus Department of Research and Development, Placental Histotherapy Center, Havana, Cuba. Department of Artificial Intelligence and Machine Learning, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Artificial Intelligence, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Electronics and Communication Engineering, Dhaanish Ahmed College of Engineering, Chennai, Tamil Nadu, India.

Abstract: Contemporary healthcare systems are confronted with escalating diagnostic complexity arising from patient heterogeneity, data fragmentation, and the cognitive limitations inherent to unaided clinical judgment. Traditional diagnostic paradigms that depend solely on physician expertise and rule-based clinical decision support systems often fail to integrate the diverse, high-dimensional information streams available at the point of care. This paper introduces MedScope AI, a novel multimodal neural diagnostic and risk forecasting framework designed to overcome these limitations by integrating three complementary data modalities: medical imaging, free-text symptom descriptions, and structured clinical records. The proposed architecture comprises a convolutional neural network (CNN) module for imaging feature extraction, a bidirectional long short-term memory (BiLSTM) encoder for natural language symptom processing, and a gradient-boosted ensemble for structured clinical data analysis. These modules are fused via a cross-modal attention mechanism that learns inter-modal dependencies and produces a unified patient representation. Experiments conducted across benchmark datasets, including NIH ChestX-ray14, MIMIC-III, and a custom-curated symptom corpus, demonstrate that MedScope AI achieves diagnostic accuracy of 94.7%, an area under the receiver operating characteristic curve (AUC-ROC) of 0.97, and a risk-prediction mean absolute error (MAE) of 0.031. The framework surpasses state-of-the-art unimodal and multimodal baselines across all evaluated metrics, establishing a new benchmark for integrated clinical decision support.

---

Keywords: Multimodal Learning; Clinical Decision Support; Convolutional Neural Networks; Cross-Modal Attention; Risk Forecasting; Medical Imaging; Natural Language Processing.

---

Received on: 03/06/2025Revised on: 10/08/2025Accepted on: 23/10/2025Published on: 07/03/2026

---

---