Smart Healthcare with AI

Diagnostic delays in resource-constrained healthcare networks remain a critical barrier to early neurological disease detection. This paper presents a Single-Stream Deep Learning Architecture that combines explainable AI with multi-modal medical image analysis for brain tumor detection. The system employs a Modified ResNet-50 with multi-modal MRI input (T1, T1c, T2, FLAIR) achieving 95.1% accuracy and 94.2% sensitivity, specifically optimized for identifying gliomas and meningiomas in limited-data environments. Critical innovation: integration of Grad-CAM heatmap generation enabling radiologist verification and trust calibration. Asynchronous processing via RabbitMQ + Celery handles ~100 medical images per minute with ~5 second response latency suitable for bandwidth-constrained tier-2 cities. Real-world validation on 500 anonymized patient records from Gorakhpur, Lucknow, and Ranchi demonstrates clinical-grade reliability with 94% radiologist agreement on AI-generated heatmaps. By addressing the explainability barrier in medical AI adoption, this work establishes a production-ready framework for trustworthy AI deployment in resource-limited telemedicine ecosystems.