Breast Cancer Detection Using Convolutional Neural Networks For MRI Image In Tanzania

Hamis, Ramadhan M (2024) Breast Cancer Detection Using Convolutional Neural Networks For MRI Image In Tanzania. Masters thesis, The Open University of Tanzania.

PDF - Submitted Version Download (16MB)

Abstract

Breast cancer is a significant global health issue, and early detection is crucial for improving outcomes. However, Tanzania faces challenges in addressing breast cancer, including a lack of locally developed Convolutional Neural Network (CNN) models. This study aims to address this gap by using CNNs with MRI images from Muhimbili National Hospital to identify breast cancer cases.The research employed an experimental study design using a dataset of 30 MRI images, with 8 malignant and 22 benign cases. To overcome data scarcity and overfitting risks, data augmentation techniques were applied, resulting in an expanded dataset of 1419 images. This augmented dataset provided a stronger foundation for training the CNN model tailored to Tanzania's context.The CNN model, developed in Python, consisted of multiple layers designed for accurate breast cancer identification. These layers included Conv2D and MaxPooling2D layers for feature capture, Dense layers for classification, and Dropout layers to prevent overfitting. The model achieved an accuracy of 96.4% and an F1 score of 96%, demonstrating its efficacy in identifying breast cancer cases.Despite the initial dataset's limitations, the research showcases the potential of CNNs and data augmentation techniques for improving breast cancer detection. Further research with larger datasets and diverse populations would be valuable for assessing the model's generalizability. Overall, this study contributes to the field of breast cancer detection by offering an efficient approach for early identification using CNNs for MRI images. Further research and validation using larger datasets and diverse populations would be valuable to assess the generalizability and scalability of the proposed model.

Actions (login required)

View Item