CLASSIFICATION OF LIVER CANCER TYPE HCC AND NON-HCC USING DEEP LEARNING

Abstract

Liver hepatocellular carcinoma (HCC) is the most prevalent form of liver cancer and a leading cause of cancer-related mortality worldwide. Early and accurate diagnosis is essential for improving patient outcomes. This study aims to develop an automated deep learning-based classification model to distinguish between HCC and non-HCC cases using portal-venous phase CT scans. Acurated dataset comprising approximately 36,000 axial slices from 390 patients (176 HCC-positive and 214 HCC-negative) was utilized, with expert-verified slice-level annotations achieving a Cohen’s κ of ≥ 0.95. To prevent data leakage, an 80/20 patient-level split was employed. Preprocessing involved converting DICOM images to PNG format, applying liver-specific windowing (W=350, L=50), resizing to 128×128 pixels, and normalization. Data augmentation techniques rotation, translation, zoom, shear, and flipping were applied to enhance model generalization. A sparse 9-layer convolutional neural network (CNN), comprising approximately 3.3 million parameters, was designed with convolutional, batch normalization, ReLU, max pooling, dense, dropout, and sigmoid layers. The model achieved a validation accuracy of 97.35%, surpassing ResNet50 (90.59%) and closely matching VGG16 (97.64%) and InceptionV3 (97.08%). It also demonstrated high diagnostic metrics, with precision, recall, and F1-score values of 0.96, 1.00, and 0.98, respectively, indicating a reduction in false negatives. These results suggest that a well-tuned CNN can achieve performance comparable to larger transfer learning models with similar computational efficiency. The findings highlight the potential of this model as a clinical decision-support tool for early HCC detection. Future work will focus on multi-center validation and integration into clinical workflows to enhance real-world applicability.