Deep learning LI-RADS grading system based on contrast enhanced multiphase MRI for differentiation between LR-3 and...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Hepatocellular carcinoma (HCC) diagnosis relies on accurate liver lesion characterization.
• Distinguishing between Liver Imaging Reporting and Data System (LI-RADS) grade 3 (LR-3) and higher grades (LR-4/LR-5) is crucial for treatment planning.
• Multiphase, contrast-enhanced (CE) magnetic resonance imaging (MRI) is a key modality for liver lesion assessment.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) method for differentiating LR-3 liver tumors from LR-4/LR-5 tumors.
• To enhance the diagnostic accuracy of hepatocellular carcinoma (HCC) detection using artificial intelligence (AI).
• To assess the performance of a convolutional neural network (CNN) in classifying liver lesions based on MRI features.

Main Methods:

• A dataset of 89 untreated LI-RADS-graded liver tumors was analyzed using multiphase 3D T1-weighted gradient echo MRI.
• Image co-registration was performed, and tumor regions were extracted for input into a pre-trained AlexNet CNN model.
• Transfer learning was applied for LI-RADS tumor grade classification, with five-fold cross-validation for model training and testing.

Main Results:

• The DL CNN model achieved high diagnostic performance, with an accuracy of 0.90, sensitivity of 1.0, precision of 0.835, and AUC of 0.95.
• The model utilized pre-contrast, arterial, and washout phases of multiphase liver MRI data.
• The CNN output probability provided radiologists with a confidence level for lesion grading.

Conclusions:

• An AlexNet CNN model demonstrated diagnostic performance comparable to expert radiologists.
• The DL approach offers valuable clinical guidance for differentiating intermediate LR-3 liver lesions from potentially malignant LR-4/LR-5 lesions.
• This AI-powered tool can aid in the accurate diagnosis of HCC.