Cross Domain Deep Neural Network For Comprehensive Health Diagnosis

Abstract

The increasing prevalence of chronic and life-threatening diseases demands rapid, accurate, and accessible diagnostic solutions. This project presents a Multi-Disease Detection System using Deep Learning-based Image Classification, integrated with a mobile healthcare application and web-based patient management dashboard for real-time disease prediction and monitoring. The proposed system is designed to detect multiple diseases including brain tumor classification (Meningioma, Glioma, Pituitary Tumor, and No Tumor), lung cancer detection (Risk of Lung Cancer / No Lung Cancer Detected), chronic kidney disease prediction (Presence / Absence of CKD), and Parkinson’s disease detection (Presence / Absence of Parkinson’s Disease) through medical image analysis. The classification framework employs a hybrid deep learning architecture combining MobileNet and Convolutional Neural Network (CNN) to enhance feature extraction efficiency and classification accuracy while maintaining lightweight deployment capability for mobile environments.

The developed Kotlin-based Android mobile application enables secure patient registration and authentication, allowing users to either upload medical images from the device gallery or capture images in real time using the mobile camera for disease prediction. Upon analysis, the application provides the predicted disease category along with recommended precautions and healthcare guidance to support early intervention. A dedicated history module stores previously uploaded scans, predictions, and timestamps for future medical reference. In addition, a web-based administrative dashboard is integrated to facilitate centralized healthcare monitoring, where administrators can manage registered patients, review uploaded diagnostic images, track prediction outcomes, and oversee disease records efficiently.

Experimental evaluation demonstrates strong performance of the proposed hybrid model, achieving an overall classification accuracy of 98.73%, with 99.52% accuracy for uploaded image predictions and 98.73% accuracy for real-time captured image analysis, indicating the robustness and reliability of the system in practical healthcare scenarios. The proposed solution offers a scalable, intelligent, and user-friendly digital healthcare platform that bridges deep learning diagnostics with mobile accessibility and centralized patient management for improved early disease detection and healthcare decision-making.