AOFNet: A novel cerebral hemorrhage segmentation network based on anatomical-omics feature

•The proposed model applies clinical data to closely mimic real-world clinical environments, offering assistance in the diagnosis of cerebral hemorrhage diseases.•An anatomical reconstruction algorithm is utilized to extract atomical-omics feature (AOF) from clinical images, facilitating a better understanding of the characteristics of clinical hematomas.•The AOFNet model incorporates an attention mechanism tailored to clinical data features and AOF correction to mitigate feature loss, while residual blocks further enhance feature mapping during the encoding-decoding process.•The model exhibits robustness in segmenting hemorrhages with irregular shapes or complex backgrounds in clinical data, and its applicability is confirmed through evaluation on publicly available datasets. Cerebral hemorrhage is a hemorrhagic cerebrovascular disease that has a rapid onset and a high mortality rate. Hemorrhage segmentation by cranial CT scan is challenging because the hemorrhagic region in the brain has a complex background and blurred boundaries with other tissues. In this paper, a novel image segmentation network based on anatomical-omics feature AOFNet is proposed, which achieves an accurate cerebral hemorrhage segmentation. The proposed network incorporates an atomical-omics feature (AOF) attentional module with an encoding-decoding sampling backbone. Using the data set of consecutive CT slices of clinical brain hemorrhage as the experimental basis, after comparing with some representative models, the proposed method in this paper performs well, with the mean values of Precision, Accuracy, and Dice of 0.945, 0.986, and 0.957, respectively. This proposed model effectively overcomes the problem of segmenting lesion that are irregular, complex in shape, and connected to the skull, which can be used as an adjunct for doctors to clinically diagnose cerebral hemorrhage and aiding in understanding the disease progression.