Breast cancer is a growing concern worldwide and has become the second leading cause of death after cardiovascular diseases. Researchers have been focusing on developing methods for early disease detection, with deep learning techniques showing promising results. However, most of these approaches have primarily focused on using breast cancer imaging, limiting the diagnosis process.
A recent study by researchers from Queen’s University Belfast and Federal College of Wildlife Management, Nigeria, aims to address this issue by proposing a novel deep learning approach that combines a twin convolutional neural network (TwinCNN) framework with a binary optimization method. This approach aims to improve breast cancer image classification by utilizing both digital mammography images and digital histopathology breast biopsy samples.
The study emphasizes the importance of a multimodal approach in medical image analysis and highlights the under-utilization of the Siamese neural network technique in previous studies on multimodal medical image classification. The proposed TwinCNN framework combines a twin convolutional neural network with a hybrid binary optimizer for feature selection and dimensionality reduction.
The TwinCNN architecture is designed to extract features from multimodal inputs, while the binary optimization method optimizes these features. Furthermore, a probability map fusion layer is introduced to fuse the multimodal images based on features and predicted labels.
The study evaluates the performance of the proposed TwinCNN framework using benchmark datasets (MIAS and BreakHis). The results show improved classification accuracy for single modalities as well as multimodality classification compared to traditional deep learning methods. The proposed binary optimizer also helps in reducing feature dimensionality and improving classifier performance.
In conclusion, the study demonstrates that the TwinCNN framework effectively addresses the challenge of breast cancer image classification by utilizing a multimodal approach. By combining image features and predicted labels, this approach improves classification accuracy and supports better diagnosis and decision-making in medical image analysis. Deep learning methods, especially those that consider multimodal data, have the potential to revolutionize early cancer detection and improve patient outcomes.