An Efficient Detection of Intracranial Hematoma Using Window-Based Stacking and YOLOv5 Framework

Traumatic brain injury (TBI) is a burgeoning medical disorder across the world particularly among young adults and children. TBI can cause intracranial hematoma (ICH), a lethal condition which requires prompt and accurate interpretation of computed tomography (CT) images for timely diagnosis and treatment. Since manual detection of CT images is a tedious and operator-dependent task, a deep learning framework is proposed for locating and categorizing ICH for improved diagnostic performance. Firstly, the input images are processed by using various techniques such as local directional pattern, local binary pattern, and windowing. Then the single stage YOLOv5 object detection model with faster spatial pyramid pooling is applied to detect hematoma regions of various types in the brain. The proposed model using windowing approach realized an overall mAP of 0.969, precision of 0.945 and recall of 0.943. The experimental results from the research study demonstrated that the suggested framework can help radiologists in strategic decision-making and improve quality care to the patients.