BrC-MCDLM: breast Cancer detection using Multi-Channel deep learning model

Breast cancer (BrC) is a lethal form of cancer which causes numerous deaths in women across the world. Generally, mammograms and histopathology biopsy images are recommended for early detection of BrC as they enable a more reliable prediction than just using mammograms. However, research indicates that even the most experienced dermatologists can detect BrC in early stage with an average accuracy of less than 80%. Over the years, researchers have made significant progress in the development of automated tools and techniques to assist radiologists or medical practitioners in BrC detection. Various machine learning and deep learning based architectures are extensively experimented on different publicly available datasets to improve the performance measures. There is further scope of improvements by extracting better representative features with deep architectural variants or ensembles techniques to minimize the misclassifications. Learnt parameters of any pretrained deep models may provide a better starting point for any other architectures using transfer learning technique. In this work, we propose computer-aided transfer learning based deep model as a binary classifier for breast cancer detection. Generally, deep learning architectures are sequential, following only a single channel for features' extraction and further classification. However, fused features extracted from multiple channels may better represent features qualitatively. The novelty of our approach is the use of multi-channel merging techniques for devising a dual-architecture ensemble. The models are trained and tested on the BreakHis dataset and an improvement in comparison with the state-of-the-arts is observed in various performance metrics. Among several combinations for ensemble architectures by utilizing various pretrained models, the Xception + InceptionV3 combination achieved an average accuracy of 97.5% for multi-channelled architecture, setting benchmarking results for further research in this direction.

Automated summary

Researchers have found great potential in deep learning technology for breast cancer detection, proposing a computer-aided transfer learning-based deep model as a binary classifier. By using multi-channel merging techniques to devise a dual-architecture ensemble, the models achieved better performance metrics compared to existing techniques.