Developing artificial intelligence to improve accuracy of radiation therapy 2.36 times higher than existing models
Nov 07, 2024
Artificial intelligence has been developed to increase the accuracy of radiation treatment plans.
Professor Park Sang-joon and Kim Jin-sung of the Department of Radiation Oncology at Yonsei Cancer Hospital announced that they have created artificial intelligence to establish a treatment plan that can increase the accuracy of radiation treatment by 2.36 times with Professor Ye Jong-cheol and Researcher Oh Yu-jin of KAIST Kim Jae-cheol of AI Graduate School.
The findings were published in the international journal Nature Communications (IF 14.7)'.
Radiation therapy is an anticancer treatment that kills cancer cells by irradiating high doses of radiation. The problem is that normal tissues around cancer cells can also be destroyed. Reducing these side effects while increasing the effectiveness of treatment is the key in establishing a treatment plan.
In order to establish a vision treatment plan, medical staff go through the process of distinguishing the contour of normal organs and cancer tissues based on patient information and imaging test results such as computed tomography (CT). There was a limitation in that it takes a lot of time as it is carried out manually.
The research team developed an AI that plans radiation therapy by using a large language model (LLM) that solves problems by learning vast amounts of data such as GPT.
The feature of the newly developed AI is that it uses a multimodal model. Through a multimodal module capable of processing various types of data such as images, audio, and video as well as text, language information in addition to image information may be reflected. Unlike the existing radiotherapy model, which relied only on imaging test results, the patient's stage, disease location, and surgical method can be additionally considered when planning.
The research team continued to verify AI performance. It boasted 1.9 times and 2.36 times higher scores than existing artificial intelligence models in external verification and expert evaluation, respectively. In an external verification to check how much the clinical target volume (CTV) selected by artificial intelligence and medical staff is matched, the research team's artificial intelligence scored 1.9 times higher than artificial intelligence using only medical images. Usually, artificial intelligence models generally have lower scores when verified with data from external institutions rather than learning institutions, but they were able to confirm excellent grades. The score was also 2.36 times better in the accuracy test evaluated by a radiologic oncologist.
Professor Park Sang-joon said, "This study is an important example of how LLM technology can be applied to actual patient care. We plan to expand the scope of AI application in the medical field through further research in the future."
Professor Park Sang-joon and Kim Jin-sung of the Department of Radiation Oncology at Yonsei Cancer Hospital announced that they have created artificial intelligence to establish a treatment plan that can increase the accuracy of radiation treatment by 2.36 times with Professor Ye Jong-cheol and Researcher Oh Yu-jin of KAIST Kim Jae-cheol of AI Graduate School.
The findings were published in the international journal Nature Communications (IF 14.7)'.
Radiation therapy is an anticancer treatment that kills cancer cells by irradiating high doses of radiation. The problem is that normal tissues around cancer cells can also be destroyed. Reducing these side effects while increasing the effectiveness of treatment is the key in establishing a treatment plan.
In order to establish a vision treatment plan, medical staff go through the process of distinguishing the contour of normal organs and cancer tissues based on patient information and imaging test results such as computed tomography (CT). There was a limitation in that it takes a lot of time as it is carried out manually.
The research team developed an AI that plans radiation therapy by using a large language model (LLM) that solves problems by learning vast amounts of data such as GPT.
The feature of the newly developed AI is that it uses a multimodal model. Through a multimodal module capable of processing various types of data such as images, audio, and video as well as text, language information in addition to image information may be reflected. Unlike the existing radiotherapy model, which relied only on imaging test results, the patient's stage, disease location, and surgical method can be additionally considered when planning.
The research team continued to verify AI performance. It boasted 1.9 times and 2.36 times higher scores than existing artificial intelligence models in external verification and expert evaluation, respectively. In an external verification to check how much the clinical target volume (CTV) selected by artificial intelligence and medical staff is matched, the research team's artificial intelligence scored 1.9 times higher than artificial intelligence using only medical images. Usually, artificial intelligence models generally have lower scores when verified with data from external institutions rather than learning institutions, but they were able to confirm excellent grades. The score was also 2.36 times better in the accuracy test evaluated by a radiologic oncologist.
Professor Park Sang-joon said, "This study is an important example of how LLM technology can be applied to actual patient care. We plan to expand the scope of AI application in the medical field through further research in the future."
|
bellho@sportschosun.com
