[1M3-CC] Machine Learning Application to Medical Engineering

[1M3-CC-02] Predictors of Intracerebral Hematoma Enlargement Using Brain CT Images in Emergency Medical Care

Kazunori Oka¹, Takumi Hirahara¹, Yasunobu Nohara² Sozo Inoue³ Koichi Arimura⁴, Koji Iihara⁵, Syoji Kobashi¹ (1. University of Hyogo, 2. Kumamoto University, 3. Kyushu Institute of Technology, 4. Kyushu University, 5. National Cerebral and Cadiovascular Center)

Intracerebral hematoma (ICH) is the cause of intracerebral hemorrhage. Acute enlargement of the ICH is high risk, and emergency surgical treatment is required. Therefore, prediction of ICH enlargement is essential to improve a survival rate and outcome. The purpose of this study is to find factors to predict the ICH enlargement with thick slice head CT images. We propose three kinds of feature extraction methods, (1) shape and texture features, (2) layered texture features, and (3) anatomical location features. In addition, we introduce an ICH enlargement prediction method using support vector machine (SVM) and feature selection. The experimental results showed that the angular second order moment of the texture feature was the most effective in predicting the ICH enlargement. By using this feature, we were able to predict the ICH enlargement with an accuracy of 75.7%. In addition, we found that normalization of the location and posture improved the prediction accuracy by 2.7% compared to that without normalization