Deep learning framework for hemorrhagic stroke segmentation and detection

This work presents a deep learning framework on Tensorflow for hemorrhagic stroke segmentation and detection from CT scans and corresponding 3D masks created by combining manual annotations with graphic morphological operations. This framework consists of three parts: data preprocessing, model training and validation. The output can be either CT image semantic segmentation results or hemorrhagic stroke detection result based on loss function selected. Our framework can be applied to various medical image segmentation and detection easily by choosing different hyperparameters. To the best of our knowledge, the present work is the first to propose a deep learning based architecture for hemorrhagic stroke segmentation, dealing with the challenges of this particular type of data. Experimental results validate the framework design and show the effectiveness of segmentation method which would significantly improve the speed and accuracy of hemorrhagic stroke detection.