Atul Kumar, Abstract Co-Author: Nothing to Disclose

Kai-Che Liu, Abstract Co-Author: Nothing to Disclose

Ching-Chun Huang PhD, Abstract Co-Author: Nothing to Disclose

Ming Hsun Lee, Abstract Co-Author: Nothing to Disclose

Lihsun Chen, Abstract Co-Author: Nothing to Disclose

Yen-Yu Wang, Presenter: Nothing to Disclose

Hurng-Sheng Wu, Abstract Co-Author: Nothing to Disclose

To automatically detect the hepatocellular carcinoma in the arterial phase CT scan of liver with the help of image gray level features in different directions (using Gabor filter with Gray Level Co-occurrence Matrix) and machine learning algorithms (using Support Vector Machine and Artificial Neural Network).

After approval from IRB, arterial phase liver CT scan image data of patients having histopathological diagnosis of hepatocellular carcinoma were retrieved from the radiology data archive of Show Chwan Memorial Hospital, Taiwan. The study was done in 125 images. Post-processing of the images was done with a median filter and an adaptive contrast enhancement technique. The images were subdivided into squares of 30x30 pixels, and based upon their content the squares were tagged as normal (liver parenchyma), tumor (hepatocellular carcinoma) and blood vessels by a radiologist. A total of 918 squares were used in the study, out of which 70% were used for training and 30% were used for test of the classification model. Directional features of the image was extracted by applying Gabor filter (a Gaussian filter function modulated by a sinusoidal plane wave) generating 18 Gabor images for each CT image. For each of the tagged region in the corresponding Gabor images, a Gray Level Co-occurrence Matrix (GLCM) based features such as energy, contrast, correlation and homogeneity were calculated. Using these features, support vector machine (SVM) and artificial neural network (ANN) classification algorithms were applied on the training squares to make mathematical classification models. The models were then applied to detect hepatocellular carcinoma in the test squares.

The sensitivity for the tumor detection was 94% with SVM and 95% with ANN classification. The overall accuracy of the classification for three different regions (tumor, vessels and normal liver) were 96% and 97% with SVM and ANN respectively.

An artificial intelligence based system for detection of hepatocellular carcinoma in liver CT was studied. The sensitivity and accuracy of the system may further improve with larger number of data.

The proposed system would be a helpful tool to physicians for automated screening for the detection of hepatocellular carcinoma.

Kumar, A, Liu, K, Huang, C, Lee, M, Chen, L, Wang, Y, Wu, H, Hepatocellular Carcinoma Detection in Arterial Phase of CT Using Directional Features and Machine Learning Algorithm.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14007064.html