DAMO Academy was selected for the representative medical AI algorithms in China in 2023

Recently, the results of the selection of representative medical AI algorithms in China for 2023 were announced. The "Hierarchical Segmentation Model of Normal Human Organs", jointly proposed by the First Affiliated Hospital of Zhejiang University School of Medicine and Alibaba DAMO Academy, was selected.

As one of the key treatment methods for cancer, the effectiveness and safety of radiotherapy highly depend on the accurate delineation of the target area and organs at risk. This requires radiation oncologists to precisely segment and delineate organs when formulating treatment plans, concentrating the radiation dose in the target area to the greatest extent possible to avoid unnecessary damage to normal organs surrounding the target area.

To address the time - consuming and labor - intensive problem of radiotherapy target area delineation, the joint research team of the First Affiliated Hospital of Zhejiang University and DAMO Academy proposed an automatic and efficient algorithm system, SOARS. They constructed a two-layer deep learning framework, achieving precise segmentation of 42 organs, making radiotherapy planning more accurate and bringing more benefits to patients. The relevant results were published in the international medical journal Nature Communications in October 2022.

The picture shows the AI organ segmentation. The red part is the radiotherapy target area.

This algorithm utilized a training set of 176 patients and was then validated with data from 1,327 patients at six external centers. The results showed that SOARS outperformed the current best international segmentation algorithms by at least 3 - 5%. It achieved delineation accuracy comparable to that of professional doctors, enabled the standardized delineation of organs at risk and normal organs, and more than 98% of the organs at risk automatically delineated by the algorithm could be directly applied in clinical practice.

Based on this algorithm, the research team further developed a segmentation model capable of continuously segmenting 143 whole - body organs. This model overcame the forgetting phenomenon in multi - organ segmentation under the condition of invisibility between datasets. The relevant achievements were accepted by ICCV 2023, a top conference in computer vision, and began to be trialed in multiple partner hospitals.

Ye Xianghua, Deputy Director of the Radiotherapy Department of the First Affiliated Hospital of Zhejiang University School of Medicine, stated that this research demonstrated that artificial intelligence can significantly reduce the workload of target area delineation, assist doctors in improving the work efficiency and repeatability of the entire cancer radiotherapy process, and is a valuable assistant for clinical doctors. "Patients often don't understand why it takes ten days to half a month to develop a radiotherapy plan. If AI can improve the efficiency in the target area delineation process, it will greatly shorten the treatment waiting time and reduce patients' anxiety."

Le Lu, Head of the Medical AI Team at DAMO Academy and an IEEE Fellow, said that this algorithm demonstrated remarkable organ segmentation performance and scalability. In the future, different departments can customize the model according to their respective needs.

This selection was reviewed by peer experts organized by the Medical Artificial Intelligence Branch of the Chinese Society of Biomedical Engineering, the Professional Committee of the Interdisciplinary Field of Mathematics and Medicine of the Society for Industrial and Applied Mathematics of China, and the Mathematics - Image Alliance on December 5, 2023. It was then reported to and approved by the Nomenclature Committee of Medical Artificial Intelligence of the National Committee for Terms in Sciences and Technologies. The awards will be presented at the 2023 China Medical Artificial Intelligence Conference at the end of December.