Intelligent Ultrasound Computed Tomography Technology (Intelligent Ultrasound CT)


Modern ultrasound CT technology typically employs ring-shaped ultrasound transducer arrays to acquire full-wave radio frequency signals (including reflection, refraction, and diffraction) and utilizes large-scale computing power for image reconstruction. Compared to traditional B-mode ultrasound which relies solely on echo signals for imaging, the grayscale values in ultrasound CT images correspond to the acoustic and mechanical properties of human tissues, facilitating the identification of early lesion morphology and the differentiation between benign and malignant tumors. Currently, the imaging speed (frame rate/real-time performance), image resolution, and detection accuracy of ultrasound CT need further improvement to meet clinical requirements. We aim to accelerate the engineering implementation of ultrasound CT technology by incorporating artificial intelligence to empower the system.

Medical Image Quality Transfer Methods


Training machine learning models enables low-quality imaging scanning equipment to "learn" how to output tissue microstructures and obtain high-quality images. However, the performance of machine learning models is often compromised by the scarcity of paired low- and high-quality clinical data and issues such as data hallucination. We investigate medical image quality transfer methods to explore high-performance and cost-effective computational approaches for generating high-quality clinical images, approaching the task from the perspective of the inverse problem of image degradation.

Data-Centric Intelligent Medical Image Analysis and Computer-Aided Diagnosis


Deploying deep learning algorithms assists doctors in "identifying" minute lesion features and improves clinical diagnostic efficiency. However, the generalization ability of models is severely constrained by issues such as multi-center data heterogeneity, high noise in labeled samples, and long-tail distributions. We research a data-centric analysis paradigm to explore pathways for the implementation of intelligent computer-aided diagnosis systems with high robustness and high trustworthiness, focusing on the perspectives of data governance and value mining.