Liver Texture Classification on CT Images of Microwave Ablation Therapy

Microwave ablation (MWA) therapy with image guidance by computed tomography (CT) is used for liver tumor destruction. However, because of the noise and therefore low contrast, CT images are not good enough for therapy control and need additional magnetic resonance imaging after the ther-apy. The ablation process itself is facing two significant chal-lenges: Firstly insufficient tumor ablation, which leads to tumor recurrence. Secondary, total ablated area was significantly larger than the tumor size which causes damaging of healthy tissue. To minimize the impact, it is crucial for the radiologist to perform the therapy well to prevent tumor recurrence. Therefore, it is essential to differentiate among healthy, tumor, and ablated tissue textures in the CT scan images. This research contributes to the understanding of tissue characterization for the reduction of the recurrence rate. In this regard, four machine-learning (ML) algorithms of Naive-Bayesian, Logistic-Regression, Decision-Tree, and Random-Forest were employed for liver tissues classification. In this paper, we propose higher order spectral particularly bispectrum analysis for extracting features from the CT images. Then classifiers were trained by ten new features extracted from the bispectrum analysis. For that, the images were divided into small patches, they were labeled as healthy, tumor, and ablated tissue. A maximum accuracy of 90.5% was obtained. The approach shows that the bispectral analysis provides valuable information that can be used during the MWA therapy for tissue characterization of CT scan even in the presence of noise.