This research presents a dual-stage deep learning system for automatic diagnosis of Soft Tissue Tumors (STS) using MRI images. The system integrates GM-UNet for tumor segmentation and EfficientNet for tumor classification. The GM-UNet model accurately isolates tumor regions using gated multi-scale attention, while EfficientNet classifies segmented tumors into benign or malignant categories. The dataset used is the Soft Tissue Sarcoma MRI collection from The Cancer Imaging Archive (TCIA). Models were trained using PyTorch in Google Colab and evaluated using Dice Coefficient, IoU, accuracy, and F1-score. A Django-based web application was developed for real-time inference, visualization, and PDF report generation. This work demonstrates a complete AI-assisted diagnostic pipeline for medical imaging applications.