A MobileNetV3-Based Approach to Breast Cancer Detection Through Transfer Learning and Histopathological Image Analysis

Breast cancer remains one of the world's most widespread and dangerous diseases, which requires the development of effective diagnostic tools. In this paper, the researchers investigate how efficient transfer learning is in breast cancer detection using the architecture of MobileNetV3. The approach of transfer learning applies knowledge gained in one task to another related task so as to enable high performance given limited data. A pre-trained MobileNetV3 large model was used, which has really very nice features extracted in large image datasets like ImageNet. The approach is to fine-tune the MobileNetV3 model with data augmentation techniques like random brightness, flip, and rotation, appended by fully connected layers to enhance prediction accuracy. On this basis, it returned a Receiver Operating Characteristic-Area Under Curve score of 0.77858 and an accuracy of 0.75979, while the loss was 0.49643. The outcome proved that with transfer learning, MobileNetV3 can become a very helpful approach to early breast cancer diagnosis and thus contribute towards better prognosis for respective patients.