Engineering A Lightweight Brain Scan Classification System through Mobilenet-Based Deep Learning

Artificial intelligence is increasingly being incorporated into healthcare applications to assist specialists in analyzing medical images more efficiently. This work presents the complete engineering process followed to build a lightweight software system capable of recognizing whether a brain magnetic resonance image contains signs of a tumor. Rather than designing a neural network from the beginning, an existing MobileNet architecture was customized and integrated into a Python-based application to reduce development time and computational requirements. Development started with organizing labelled MRI images into separate categories before preparing them for machine learning. Each scan was resized, converted into a uniform numerical format, and supplied to the learning framework for optimization. The implementation relied on TensorFlow for model development, NumPy for numerical processing, and scikit-learn for evaluating prediction quality. All software components were created and tested using Visual Studio Code. After optimization, the trained network was validated using images that were excluded from the learning process. The completed application achieved an overall prediction accuracy of 97.3%, demonstrating that compact pretrained architectures can provide reliable image classification without requiring specialised computing hardware. The developed software establishes a practical foundation for future expansion involving multiclass diagnosis, cloud deployment, and explainable artificial intelligence.