Search

Cerebral stroke is a severe health issue that necessitates prompt and intensive medical care. Identifying the location of any hemorrhage present is a critical step in treatment. This study compares the segmentation accuracy and precision of various fully convolutional neural network (FCN) methods applied to cerebral stroke detection. An open public dataset containing 318 paired 2D brain CT and stroke labeling images was used to investigate the performance of different FCN methods. The Xception, InceptionResNetV2, MobileNetV2, ResNet50, and ResNet18 FCN models were employed to segment stroke locations in CT images. The hyperparameters included batch size, epoch size, and learning rates (0.001). The dataset was divided into 70%, 20%, and 10% for training, validation, and testing of the FCN models, respectively. Segmentation performance was assessed using intersection over union (IoU) and dice score metrics. The InceptionResNetV2 FCN approach, in combination with an SGDM optimizer, a batch size of 3, and an epoch size of 15, achieved an IoU of 0.632 and a dice score of 0.610 in the testing set. This study investigated the performance of five FCN approaches with various hyper-parameters on CT stroke images. The InceptionResNetV2 FCN method demonstrated potential usefulness for segmenting cerebral stroke locations in CT images. Future work should consider using a larger CT dataset to further validate this approach for clinical applications. [ABSTRACT FROM AUTHOR]