CNN architectures
SMOTE
Data segmentation
Deep Learning
- A comperative study of CNN, fine-tuned VGG16 and ResNet50 architectures for breast tumor detection from BUS-BRA dataset of ultrasound images, achieving classification accuracy of 85.2%, 91.6% and 93.4% respectively
- Leveraged U-Net for segmentation, SMOTE for class imbalance mitigation and data augmentation techniques, to significantly enhance generalizability of the model on a dataset of 1875 annotated ultrasound images