(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: The aim of this study is to evaluate diagnostic performance of a deep learning scheme on detecting liver cirrhosis having as input ultrasound (US) B-Mode images. Segmented liver left lobes were delineated by an expert radiologist, on patients with Chronic Liver Disease (CLD) using liver biopsy (LB) as the ‘Gold Standard’.
Methods: 69 consecutive CLD diagnosed patients (14 F0-F3, and 55 F4) underwent a regular abdomen US B-Mode and LB examination using Metavir scale (F0-F4). A B-Mode image containing liver left lobe was acquired by an expert Radiologist performing the US examination. Then, the radiologist delineated the boundaries of liver surface for each patient on the saved image to acquire a binary image as liver’s left lobe mask (pixels of liver left lobe having values of ‘1’ and pixels out of the liver having values of ‘0’). The mask images were fed to GoogLeNet pre-trained deep learning scheme using LB fibrosis staging (F0-F4) as labels for fine tuning. Data were separated in two classes according to their steatosis grade: F0-F3 (Non-cirrhotic) as class 1 and F4 (Cirrhotic Class) as class 2. The mask images were separated in training (70%) and testing (30%) and corresponding training and testing accuracies were calculated. The random image separation and GoogLeNet fine tuning process was repeated 30 times.
Results: The deep learning scheme’s mean accuracy reached 100% for the training data while the mean accuracy of the trained model on the test data reached 91%.
Conclusion: The results indicate that this scheme can be used as a supplementary tool for cirrhosis detection on regular US B-Mode abdomen examination.
Keywords
CAD, Quantitative Imaging, Ultrasonics
Taxonomy
IM- Ultrasound : Machine learning, computer vision
Contact Email
