[Hinews / HEALTH IN NEWS] SEOUL — A team of radiologists at Incheon Hi Hospital, led by Professors Jeongho Lee, Jiwoon Seo, Hyojin Kim, Jihyun Lee, and Jiwon Chae, has developed an artificial intelligence (AI) model that significantly improves early detection of metastatic cancer through spinal MRI analysis. The model is expected to enhance diagnostic accuracy and reduce the workload of medical professionals.

The study utilized spinal MRI data from 322 patients collected between 2015 and 2021 to train a deep learning-based AI model, U-Net, designed to detect and segment metastatic lesions. The AI system integrates three types of MRI imaging modalities and was evaluated against the diagnostic performance of five radiology specialists.

Professors Donghyun Kim (left) and Jiwoon Seo (right), Department of Radiology, Seoul Metropolitan Government Boramae Medical Center (Photo courtesy of Seoul Metropolitan Government Boramae Medical Center)
Professors Donghyun Kim (left) and Jiwoon Seo (right), Department of Radiology, Seoul Metropolitan Government Boramae Medical Center (Photo courtesy of Seoul Metropolitan Government Boramae Medical Center)
Results showed the AI model achieved an accuracy of 85.7% in identifying metastatic lesions, surpassing the specialists’ average accuracy of 74.6%. Notably, the AI excelled at detecting lesions smaller than 1 centimeter, demonstrating its potential for early diagnosis. The model also exhibited a lower rate of false positives compared to human specialists.

Bone metastases, common in patients with lung, breast, and prostate cancer, frequently affect the spine, potentially causing pathological fractures or nerve damage. Early detection is critical, but manually reviewing hundreds of MRI images places a significant burden on radiologists. The AI’s diagnostic support is poised to become a valuable tool in clinical settings.

“This study is significant because the AI demonstrated diagnostic accuracy comparable to that of specialists,” said Professor Jeongho Lee. “It lays an important foundation for advancing and practicalizing AI-based diagnostic systems in the future.”
The research was published in the 2024 issue of the Korean Journal of Radiology, with full details available in the original paper.

저작권자 © Hinews 무단전재 및 재배포 금지
ad