The primary objective of this project is to enhance the accuracy and efficiency of breast cancer detection using ultrasound images by leveraging both weakly supervised and fully supervised learning models. To achieve this, we utilized pre-trained CNN models such as VGG19, MobileNet, and ResNet50. This approach aims to address the data annotation challenges by exploring alternative supervised learning techniques.

Data Collection and Preprocessing:

Model Selection:

Weakly Supervised Learning:

Fully Supervised Learning:

Training and Validation:

Evaluation and Comparison:

Fully supervised models outperformed weakly supervised models, demonstrating superior accuracy and robustness in breast cancer detection.

Among tested models, fully supervised ResNet50 showed the highest performance, significantly surpassing VGG19 and MobileNet.

This project showcases the effectiveness of fully supervised learning for breast cancer detection using ultrasound images. Integration of pre-trained CNN models like VGG19, MobileNet, and ResNet50 provides a robust solution, with fully supervised ResNet50 delivering the best diagnostic accuracy. Future work will focus on optimizing these models further and exploring semi-supervised and unsupervised learning techniques to reduce dependency on annotated datasets.

Feel free to adapt and expand upon this README template for your GitHub repository.