MBFormer: Microbubble Transformer for 3D Time-Series Data Processing to Improve Bound Bubble Detection in...

Development of nondestructive ultrasound molecular imaging (UMI) is essential for early cancer detection through real-time, free-hand screening using clinical ultrasound systems. Current techniques face challenges in accurately detecting targeted microbubbles (MBs) bound to specific biomarkers, primarily due to false-positive detections of unbound free-floating MBs. We propose a transformer model designed for time-series data processing, aimed at improving the differentiation of bound MBs in UMI. We propose a hierarchical transformer, termed "MBFormer" (microbubble transformer), featuring a positional-embedding-free encoder and a lightweight decoder. This model leverages attention mechanisms within 3D spatio-temporal data to effectively capture stationary signals from bound MBs while suppressing non-stationary signals from unbound MBs. Given that MBs manifest as relatively small textures compared with conventional segmentation targets in medical imaging, such as organs and tumors, we optimized the model with two hierarchical layers, each with an attention block, to process spatio-temporal (ultrasound video) data. The network outputs the molecular signal amplitude to visualize fine MB textures. Performance evaluation was conducted using an in vivo animal model of breast cancer, with comparisons made against a prior convolutional neural network (CNN)-based UMI method and SegFormer3D, a representative 3D transformer backbone. The results indicate that MBFormer (AUC = 0.943) outperformed both CNN (AUC = 0.897) and SegFormer3D (AUC = 0.766) in detecting bound MBs. The CNN exhibited residual molecular signal from free-floating MBs in the cardiac chambers, whereas SegFormer3D baseline failed to detect fine MB textures effectively. Overall, MBFormer demonstrated enhanced detection of bound MBs while suppressing free-floating MB signals. Furthermore, MBFormer achieved a frame rate of 16.7 to 18.1 frames per second, demonstrating its potential for real-time application. We anticipate that this improved transformer-based UMI model can facilitate real-time, free-hand nondestructive UMI in clinical systems.