LeViT–ResNet hybrid AI model for transferable early detection and classification of liver cancer
Year: 2026
Type: article
Abstract: Accurate staging and classification of liver fibrosis from ultrasound images remains a complex challenge due to low contrast, noise artifacts, and heterogeneous tissue patterns. The staging of liver fibrosis is the main focus of this study. Ultrasound images are categorized into stages of fibrosis severity, and segmentation is only utilized to isolate liver regions related to fibrosis, not to identify or categorize tumors. This paper presents a novel, transferable AI framework that combines a Savitzky-Golay refinement filter, an Adaptive Seed Region Growing (ASRG) segmentation algorithm, and a hybrid LeViT–ResNet-based classifier. The feature extraction phase leverages contour-based shape descriptors to represent morphological complexity, while the classifier architecture integrates local (CNN-based) and global (Vision Transformer-based) cues using Residual Vision Transformer (ResViT) blocks. To fine-tune hyperparameters, the Sunflower Optimization Algorithm (SFO) is applied, ensuring robust convergence and optimal learning dynamics. Evaluation on a publicly available liver ultrasound dataset demonstrates that the proposed method outperforms state-of-the-art models such as MultiPatch-ViTNet and MultiPatch-ViT-artificial neural network (ANN) in terms of accuracy, AUC, F1-Score, and class-wise discrimination, particularly for intermediate fibrosis stages. These results confirm the model’s potential for real-world deployment in non-invasive liver disease screening and diagnosis.
Source: Scientific Reports
Authors: B. Deepa, C. N. Marimuthu
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7225898990
