Jump to the main content block
:::

Medical Imaging Research: Professor Wang Ching-Wei Develops Universal 3D Lesion Segmentation AI Model [August 2024]

Professor WANG Ching-Wei, a leading expert in medical imaging technology and director of Taiwan Tech's Graduate Institute of Biomedical Engineering, has developed a new AI model that rapidly and accurately identifies and segments various types of lesions in CT scan images. Her team secured 3rd place in the 2024 International Medical 3D CT Image AI Competition, the Universal Lesion Segmentation '23 Challenge.

臺科大醫工所王靖維教授(右)團隊於今年國際醫療3D CT影像AI競賽,榮獲第3名佳績。圖左為王靖維教授指導學生醫工所碩二蘇鼎盛。
Professor WANG Ching-Wei (right) from Taiwan Tech's Institute of Biomedical Engineering, with master’s student Ting-Sheng Su (left).

The Universal 3D Lesion Segmentation AI Model, developed by Professor Wang and her team, precisely identifies lesions affecting the thorax and abdomen, including those in bones, the pancreas, kidneys, liver, colon, and lymph nodes. The model is tailored for thoracoabdominal CT images, assisting radiologists in 3D lesion annotation and addressing the issue of high labor costs associated with manual annotation.

王靖維教授團隊開發通用AI模型精準3D分割多種類別病灶,紅色輪廓代表基準真相;綠色輪廓代表模型預測。
3D lesion segmentation results: The red contour represents the baseline truth, the green contour represents the model's prediction.

Automated AI lesion segmentation for CT scans offers significant advantages over manual segmentation, such as improved efficiency, repeatability, accuracy, and standardization, leading to more precise quantitative analysis. Traditional manual annotation takes about 30 to 60 minutes per case, whereas Professor Wang’s model can process each 3D lesion in just 3.25 seconds [on the Grand Challenge platform server equipped with a single T4 GPU, and less than 2 seconds on a computer equipped with an RTX 4080].

臺科大醫工所王靖維教授團隊於ULS23競賽,榮獲第3名佳績。
Professor WANG and her team are proud of their third place at 2024 International Medical 3D CT image AI competition.

The team stood out among 632 participants in this year's Universal Lesion Segmentation '23 Challenge [ULS23], earning 3rd place. The ULS23 competition, held on the Grand Challenge platform, aims to promote research on universal lesion segmentation models in the 3D CT field. The competition provided a clinical test set of 39,500 3D CT lesion images, allowing participants to build and validate multi-category thoracoabdominal lesion models.

Video: https://www.youtube.com/watch?v=kPQLhCx9ViA
Contact: 
Professor WANG Ching-Wei 
 Director of the Graduate Institute of Biomedical Engineering

Email: cweiwang@mail.ntust.edu.tw 

 

 

 

Click Num: