Research

Mobile ultrasound devices play a crucial role in emergency and family doctor services, providing valuable diagnostic capabilities in a portable form factor. However, when compared to traditional medical ultrasound equipment, there is still scope for improvement in terms of image quality.

We propose the Hyper-CycleGAN model as a solution to enhance the image quality of mobile ultrasound devices. This model aims to establish a structure-preserving mapping between low-quality ultrasound images and their corresponding high-quality counterparts. Additionally, we plan to incorporate an attention mechanism into the model to enable it to better focus on important areas within the image, thus further enhancing the diagnostic relevance of the generated high-quality images. By leveraging advanced techniques and algorithms, our approach aims to bridge the gap in image quality between mobile ultrasound devices and traditional medical ultrasound equipment, thereby improving the accuracy and effectiveness of emergency and family doctor services.

The study of breast dense tissue holds immense significance in the evaluation of breast health, early detection of breast diseases, and risk assessment of breast cancer. Breast density refers to the presence of a higher proportion of glands and connective tissue in the breast, relative to fat tissue. Research has indicated that women with denser breast tissue have a higher risk of developing breast cancer compared to those with a higher proportion of fat tissue. Therefore, investigating dense breast tissue plays a crucial role in enabling doctors to more accurately assess breast health. Our research primarily focuses on employing various techniques to segment Magnetic Resonance Imaging (MRI) images of breasts and identify dense tissues, in order to calculate breast density. This enables initial screening and assessment of breast health, as well as evaluating the risk of breast cancer.

The main challenge in breast cancer event prediction is how to capture dynamic breast tissue changes while accurately modeling the chronological order and interval between events. In addition, data sparsity and inadequate labeling further complicate this task, and improving the interpretability of predictive models is particularly critical to help physicians make early diagnoses and personalize treatment. Therefore, an approach called OA-BreaCR has been proposed that focuses on integrating dynamic changes and time series data of longitudinal breast tissue through a long-term attention alignment model. It aims to address complex temporal alignment, data sparsity, and interpretability issues in breast cancer event prediction, providing a reliable tool for early diagnosis and personalized treatment, while improving the accuracy and practicality of prediction. By accurately predicting the timing of future breast cancer events, the OA-BreaCR model improves prediction accuracy and provides healthcare workers with transparent, interpretable results that can help drive early intervention and personalized treatment, while effectively addressing data sparring and optimizing available resources.

Multi-tasking Breast Density Classification

Multi-tasking Breast Density Classification

Multimodal breast cancer risk prediction