In the field of computer-aided diagnosis， accurate segmentation of IntraCranial Hemorrhages （ICHs） in Computed Tomography （CT） images is crucial for subsequent treatment and prognosis. To address the challenge of segmenting small hemorrhage regions， an information bottleneck-guided ICH segmentation method was proposed. Based on this method，an Information Bottleneck-Guided Segmentation Network （IBGS-Net） was proposed. Firstly， the U-Net architecture was used as a base， and an information bottleneck layer was introduced to enhance the recognition of key features related to ICH segmentation. Then， through the designed Residual SPatially-ADaptivE normalization （ResSPADE） module， the Information Activation Map （IAM） was integrated effectively into the segmentation process， thereby improving the network’s ability to identify and locate hemorrhage regions. Finally， an Interactive Guided Loss （IGL） function was introduced to optimize the model’s processing of difficult-to-segment regions， thereby further enhancing the model’s generalization performance. Evaluation results on the internal dataset indicate that the proposed method achieves 78.1% in Dice Similarity Coefficient （DSC）， 90.1% in Normalization Surface Dice （NSD）， and 11.5% in Relative Volume Difference （RVD）. On the public dataset INSTANCE 2022， the results of comparison with other segmentation methods show that compared to the suboptimal results， the three indicators of the proposed method increased by 1.9， 2.4， and decreased by 3.2 percentage points， respectively. The above validates the effectiveness and superiority of the proposed method in ICH segmentation tasks， so that the method is suitable for assisting clinicians in segmenting ICHs.