News

Oncosoft develops radiation cancer treatment technology with LLM for greater accuracy

[MEDIGATE NEWS] Lim Sol  |
2024-11-08
(From left) Prof. Park Sang-Joon, Yonsei Cancer Hospital, Dr. Oh Yoo-Jin, Massachusetts General Hospital, USA, and
Kim Jin-Sung, CEO of Oncosoft
(Photo provided by Oncosoft)

Professors Park Sang-Joon and Ye Jong-Cheol, Department of Radiation Oncology, Yonsei University Cancer Center, and Dr. Yoo-Jin Oh, Massachusetts General Hospital, USA


Oncosoft announced today that it has developed an AI-based tumor contouring technology that can increase the accuracy of radiation therapy by 2.36 times in collaboration with Prof. Sangjun Park and Jongchul Yeh from the Department of Radiation Oncology at Yonsei Cancer Hospital and Dr. Yoojin Oh from Massachusetts General Hospital (MGH) in the United States. The findings were published in the international journal Nature Communications.

Radiotherapy is a method of killing cancer cells by irradiating them with high doses of radiation, but it comes with the risk of damaging the surrounding normal tissue. Therefore, it is important to develop a precise treatment plan that can effectively target cancer cells while protecting normal tissue. To this end, OncoSoft has developed OncoStudio, an AI-based automatic contouring software that utilizes imaging data such as CT to distinguish the outlines of normal organs around cancerous tissue to support radiotherapy planning.

According to the company, by building a multimodal AI model based on a large language model (LLM) such as GPT, it has become possible to establish a treatment plan that reflects language-based clinical information such as the patient's stage, disease location, and surgical method in addition to existing image information.

As a result of the performance verification of the joint research team, the external evaluation and expert evaluation consisting of radiation oncologists scored 1.9 times and 2.36 times higher than the existing AI model, respectively. In particular, the external validation, which evaluated the matching of tumor volume (CTV), the target of treatment, showed 1.9 times higher accuracy than the existing model.

The researchers concluded, “Multimodal AI has the advantage of being able to segment the location and shape of the tumor more accurately using clinical information, although it is difficult to accurately determine left or right breast cancer based on images alone. We have seen similar results in targeting radiation therapy for prostate cancer as well as breast cancer.”

Source=Nature. Comparison of an AI model that utilizes traditional video alone and a multimodal AI that utilizes LLM and video together.

"The new AI technology that reflects not only image data but also text-based clinical information can further improve the precision and safety of radiotherapy planning," said Jin Sung Kim, CEO of Oncosoft. "In addition to the normal organ contour contouring technology provided by OncoStudio, tumor volume (CTV) contouring is now possible, and the results of multi-center collaborative research with overseas institutions will open a new path for radiation cancer treatment."

https://medigatenews.com/news/2732995437