Recently, the research team led by Professor Zhuang Zhemin, from the Guangdong Provincial Engineering Technology Research Center of Artificial Intelligence and Modern Ultrasound and the Key Laboratory of Intelligent Manufacturing under the Ministry of Education, has successfully achieved the transformation of patented achievements with a transaction value of 1,050,000 yuan. This achievement not only represents high recognition of the team’s scientific research strength but also marks a crucial step for their research in the interdisciplinary field of artificial intelligence and ultrasound engineering—moving from the laboratory to industrial application.

Professor Zhuang Zhemin’s team has long been engaged in the integrated innovation of artificial intelligence and ultrasound imaging technology. The three patents transferred this time, focusing on deep learning‑based volumetric ultrasound image processing, are important outcomes of the National Natural Science Foundation project Research and Application of Breast Tumor Segmentation and Classification Models Based on Fully Automated Breast Volumetric Ultrasound Imaging and 3D Deep Learning led by Professor Zhuang. The licensed patents have been applied in a domestically pioneering fully automated breast volumetric ultrasound imaging system jointly developed by the partner company and the research group. They enable multi‑level classification of breast tumors, segmentation of tumor lesions and contrast‑enhanced regions, and layered analysis of breast tissue using deep learning models, opening up new avenues for the application of artificial intelligence in female breast cancer screening.

“It has taken us nearly a decade to refine technologies from laboratory prototypes to mature solutions recognized by industry,” said Professor Zhuang Zhemin. The team has always adhered to the research philosophy of demand‑oriented and application‑targeted. The partner company chose to cooperate precisely for the innovation and practicality of the technologies. In the future, both parties will conduct in‑depth cooperation on technological iteration and scenario expansion, promoting the solid landing of scientific research achievements and injecting new momentum into industrial upgrading.

Alongside the patent transformation, the team has also achieved remarkable academic progress. Two recent research papers were published in internationally authoritative journals:

These publications demonstrate the team’s cutting‑edge exploration in artificial intelligence and medical imaging applications.

The paper published in Medical Image Analysis, titled Multimodal sparse fusion transformer network with spatio-temporal decoupling for breast tumor classification, integrates multimodal imaging including ultrasound, superb microvascular imaging, and strain elastography through spatial–temporal feature decomposition. Using a sparse cross‑attention module, it selectively enhances interactions between different modalities to reduce feature redundancy and inter‑modality discrepancies, facilitating more effective multimodal fusion. A mixed‑scale convolution module then captures multi‑resolution anatomical and textural features to achieve tumor classification. The method delivers higher clinical reliability and computational efficiency, effectively supporting intelligent diagnosis in breast ultrasound imaging.

The paper in Expert Systems With Applications, EMCANet: An edge-refined multi-scale convolutional hybrid attention network for robust tumor segmentation in automated breast ultrasound, proposes a novel and efficient edge‑refined multi‑scale convolutional hybrid attention network (EMCANet) for automatic tumor segmentation in automated breast ultrasound (ABUS). The model first extracts edge features along tumor contours via an edge refinement and fusion module, then captures multi‑scale contextual clues through a spatial multi‑scale convolutional hybrid attention module. Finally, group‑wise correlation modeling aligns multi‑stage features, enabling robust segmentation of tumors with varying sizes and morphologies in ABUS images. This approach helps clinicians reduce the time required for tumor segmentation assessment and lower the risk of missed lesions.

As a core scientific research team in artificial intelligence and intelligent manufacturing at Shantou University, the group has long adhered to a development model of deep integration of industry, university, research, and application, forming a virtuous cycle:

basic research → technological breakthrough → achievement transformation → feedback to research

The team also attaches great importance to talent training and has established a hierarchical research training system covering PhD students, master’s students, and undergraduate students. Many students have not only improved their academic abilities but also gained rich practical experience through research projects. In recent years, graduates from the team have joined leading enterprises such as United Imaging Healthcare, Mindray, and Huawei for technical research and development, or pursued further studies at domestic and foreign universities, providing a large number of high‑quality talents for the industry.

In the future, the university will further strengthen support for scientific research teams, improve the collaborative innovation mechanism of industry‑university‑research‑application, and encourage more researchers to step out of the laboratory and turn scientific achievements into real productivity for social development.

Photos & Text: School of Engineering