Automated Intracerebral Hemorrhage Detection in CT Images Using InceptionResNetV2

Intracerebral hemorrhage (ICH) is a common cerebrovascular disease with a high mortality rate, making accurate and timely identification crucial. This paper introduces a machine learning-based model for detecting ICH from brain CT images, aimed at improving the automation and accuracy of medical image diagnostics. The model leverages a InceptionResNetV2 as the feature extraction network, combined with global average pooling, dropout, and fully connected layers tailored for binary classification tasks. Experimental results show that the model achieves an accuracy of 0.9733 on the validation set and 0.9200 on the test set. The loss rates are 0.1644 and 0.1567, recall rates are 0.9864 and 0.9623, F1 scores are 0.9638 and 0.9273, and AUC values are 0.9972 and 0.9889 for the validation and test sets, respectively. These metrics indicate the model's strong generalization capability. Additionally, a user interface was developed to predict the probability of ICH in brain CT images, providing an effective tool for automated detection. This tool holds promise in assisting clinicians with rapid and accurate diagnostic decisions.