応用生物情報学および計算生物学ジャーナル

Hemorrhage Detection Using Transfer Learning

Pokala Pranay Kumar*, Gayathri Kalva, Raul Villamarin Rodriguez, Samala Nagaraj

Bleeding, likewise named hemorrhage, is the name used to depict blood loss. It refers to blood clots inside the body or outside the body, which are called internal or external bleeding. Blood clots can take place anywhere in the body. Internal bleeding usually appears when there is a blood leak seen through a damaged blood vessel or an organ. When bleeding is seen to be taking place inside the skull it is known as Intracranial Hemorrhage. This leads to various causes like head trauma, bleeding tumours, high blood pressure, blood clotting disorders, and others. Neurosurgeons use a Computed Tomography (CT) to detect hemorrhage which appears to be intense (white) and try to identify its subtypes by extracting some essential features depending upon the shape, location, and vicinity. Deep Learning can change the future of health care. Using Deep Learning Algorithms, we can try analyzing the data from electronic documents like CT Scans or X-Rays. The aim is to classify the hemorrhage images; the results states it will detect the hemorrhage from the images. In this, we are using deep learning methods like transfer learning where we are training with predefined networks like alexnet, googlenet, resnet, etc. For this, we used Densenet as our pretrained network which was trained and tested with 200 brain CT images. The resultant accuracy on the training dataset is 99.38% and the test is 94.44%.