SDR-Former: A Siamese Dual-Resolution Transformer for Liver Lesion Classification Using 3D Multi-Phase Imaging

TL;DR

SDR-Former tackles automated liver lesion classification across 3D multi-phase CT and MR images with varying phase counts by introducing a Siamese weight-sharing backbone, a hybrid Dual-Resolution Transformer (DR-Former), a Bilateral Cross-resolution Integration Module (BCIM), and an Adaptive Phase Selection Module (APSM). The architecture jointly leverages high-resolution CNN detail and low-resolution Transformer context, enabling effective cross-phase interactions and phase-aware weighting, which yields state-of-the-art performance on a three-phase CT dataset and an eight-phase MR dataset. Extensive ablations confirm the DR-Former’s advantage over single-branch networks, and the BCIM/APSM modules provide additional gains, while cross-modality transfer learning and versatility on MedMNIST3D demonstrate broad applicability. The work also contributes a publicly released multi-phase MR liver dataset to support future research and MICCAI LLD-MMRI challenges, highlighting practical impact in scalable, transfer-friendly multi-phase medical imaging analysis.

Abstract

Automated classification of liver lesions in multi-phase CT and MR scans is of clinical significance but challenging. This study proposes a novel Siamese Dual-Resolution Transformer (SDR-Former) framework, specifically designed for liver lesion classification in 3D multi-phase CT and MR imaging with varying phase counts. The proposed SDR-Former utilizes a streamlined Siamese Neural Network (SNN) to process multi-phase imaging inputs, possessing robust feature representations while maintaining computational efficiency. The weight-sharing feature of the SNN is further enriched by a hybrid Dual-Resolution Transformer (DR-Former), comprising a 3D Convolutional Neural Network (CNN) and a tailored 3D Transformer for processing high- and low-resolution images, respectively. This hybrid sub-architecture excels in capturing detailed local features and understanding global contextual information, thereby, boosting the SNN's feature extraction capabilities. Additionally, a novel Adaptive Phase Selection Module (APSM) is introduced, promoting phase-specific intercommunication and dynamically adjusting each phase's influence on the diagnostic outcome. The proposed SDR-Former framework has been validated through comprehensive experiments on two clinical datasets: a three-phase CT dataset and an eight-phase MR dataset. The experimental results affirm the efficacy of the proposed framework. To support the scientific community, we are releasing our extensive multi-phase MR dataset for liver lesion analysis to the public. This pioneering dataset, being the first publicly available multi-phase MR dataset in this field, also underpins the MICCAI LLD-MMRI Challenge. The dataset is accessible at:https://bit.ly/3IyYlgN.

Figures (7)

• Figure 1: This diagram illustrates the SDR-Former framework, which utilizes a Siamese Neural Network (SNN) for processing each phase in the multi-phase imaging series independently yet simultaneously. Central to this framework is the hybrid Dual-Resolution Transformer (DR-Former), a fusion network that effectively merges CNN and Transformer approaches to handle high- and low-resolution inputs using shared weights. The Bilateral Cross-Resolution Integration Module (BCIM) is strategically incorporated into this architecture to facilitate semantic interactions and enhance the coalescence of the distinct high- and low-resolution streams. Furthermore, the Adaptive Phase Selection Module (APSM) plays a key role in dynamically merging these feature representations, ensuring a versatile and perceptive amalgamation of multi-phase features.
• Figure 2: A diagrammatic depiction outlining the structure of the Hybrid Dual-Resolution Transformer (DR-Former).
• Figure 3: A schematic illustration of the Bilateral Cross-resolution Integration Module (BCIM).
• Figure 4: A schematic illustration of the Adaptive Phase Selection Module (APSM).
• Figure 5: Receiver Operating Characteristic (ROC) curves for various methods.