A retrospective systematic study on hierarchical sparse query transformer-assisted ultrasound screening for early hepatocellular carcinoma

Hepatocellular carcinoma (HCC), ranking as the third leading cause of cancer-related mortality worldwide, demands urgent improvements in early detection to enhance patient survival. While ultrasound remains the preferred screening modality due to its cost-effectiveness and real-time capabilities, its sensitivity (59%–78%) heavily relies on radiologists’ expertise, leading to inconsistent diagnostic outcomes and operational inefficiencies. This study introduces the Hierarchical Sparse Query Transformer (HSQformer), a novel hybrid architecture that synergizes Convolutional Neural Networks’ (CNNs’) local feature extraction with Vision Transformers’ (ViTs’) global contextual awareness through latent space representation and sparse learning. Unlike prior approaches that often require complex structural modifications or sophisticated design techniques, HSQformer dynamically activates experts via a Mixture-of-Experts (MoE) framework, achieving hierarchical feature integration without structural redundancy. Evaluated across three clinical scenarios: single-center, multi-center, and high-risk patient cohorts, HSQformer outperforms state-of-the-art (SOTA) models and matches senior radiologists’ diagnostic accuracy while significantly surpassing junior counterparts. These results demonstrate HSQformer’s potential to standardize HCC screening, reduce human expertise dependency, and improve early diagnosis rates. The full code is available at https://github.com/Asunatan/HSQformer .