Akademik Veri Yönetim Sistemi
International Conference on Computing, Intelligence and Data Analytics, ICCIDA 2022, Kocaeli, Türkiye, 16 - 17 Eylül 2022, cilt.643 LNNS, ss.304-318
Brain hemorrhages are one of the most dangerous disease groups. If not detected early, it can lead to death or severe disability. The most common method used to detect bleeding is the evaluation of computed tomography (CT) images belonging to the bleeding area by specialist physicians. Considering the difficulty of access to neurosurgery specialists and the lack of expertise of other doctors in emergency intervention on the subject, there is a need for decision support mechanisms to assist physicians in the diagnosis and treatment process. Artificial intelligence-based systems to be used for this purpose can accelerate the diagnosis and treatment process while reducing the burden on physicians. In this study, the suitability of Mask Region-Based Convolutional Neural Network (Mask R-CNN), Cascade Region-Based Convolutional Neural Network (Cascade R-CNN), Mask Scoring Region-Based Convolutional Neural Network (MS R-CNN), Hybrid Task Cascade (HTC), You Only Look At Coefficients (YOLACT), Instances as Queries (QueryInst), and Sample Consistency Network (SCNet) methods, investigated for the problem of detection and segmentation of subdural brain hemorrhages. The performance of the methods was determined over the images in the CQ500 dataset. This is one of the few studies that perform segmentation of subdural cerebral hemorrhages using CT images from an open dataset. The results were evaluated according to Intersection Over Union (IoU) and Mean Average Precision (mAP) metrics. Experimental results showed that two methods could detect and segment subdural hemorrhages more accurately than the others.