Leveraging Pre-trained Models: A Comparative Analysis of Transfer Learning Strategies for medical image analysis

This role is critical for the early diagnosis and detection of diseases, among them cancer, through the analysis of medical images. In this paper, different transfer-learning approaches are compared regarding the task of breast cancer detection using deep pretrained Convolutional Neural Networks. Custom model CNN and transfer learning using VGG16 architecture were implemented. Such a dataset consists of images of breast cancers labeled as healthy, 0, and cancerous, 1. Finally, all this work integrates a performance metric evaluation that enables Gradio for the prediction of the real-time interactive user interface. Our results show that this fine-tuning significantly increased accuracy with less training time compared to custom models; hence, it may be useful for extensive medical image classification tasks.