Combined regional and spatio-temporal approach improves hepatic tumors classification in Multiphase CT

In this work, we investigate the effect of using spatio-tepmoral features on a regional basis on the liver focal lesions classification performance in the multiphase CT images. Texture, Density, and temporal feature set and their different combinations along spatial partitioned ROI were investigated to better characterizing five hepatic pathologies from multiphase contrast-enhanced CT scans. Embedded feature selection followed by decision tree ensembles classification with ten folds cross-validation were employed to classify a total of 180 ROI includes normal tissues, cyst, haemangioma, metastatic and hepatocellular carcinoma. Our result suggested that normal liver tissues could easily be recognized from just the density features, whereas texture features could obtain near best results in classifying HCC. Combining all feature sets could overcome individual performance variations between them and attain consistent better results for all tumor types. Moreover, Adding the regional information improves all the classes characterization especially haemangiomas and metastases.