Multiple abnormality detection for automatic medical image diagnosis using bifurcated convolutional neural network

Convolutional Neural Networks (CNNs) are widely adopted in the automatic analysis of abnormalities in medical imaging applications. Although CNN structures are robust feature extractors, there are two problems. First, simultaneous classification and segmentation of abnormal regions in medical images is a crucial and difficult task. Second, the implementation of CNN’s with high computational complexity in portable devices imposes power and resource constraints. To address these problems, we propose a bifurcated structure with one branch performing classification, and the other performs the segmentation. Initially, separate network structures are trained for each abnormality separately and then primary parts of these networks are merged. The bifurcated structure has a shared part, which works for all abnormalities. One branch of the final structure has sub-networks for segmentation of different abnormalities, and the other branch has separate sub-networks, each is designed for classification of a specific abnormality. Results of the classification and segmentation are fused to obtain the classified segmentation map. The proposed framework is simulated using four frequent gastrointestinal abnormalities as well as three dermoscopic lesions. Properties of the bifurcated network such as low complexity and resource sharing, make it suitable to be implemented as a part of portable medical imaging devices.