Analyzing medical image data has always been a complex and tedious process. Recently, researchers at Weill Cornell Medicine developed LILAC (Learning-based Inference of Longitudinal Image Changes), a novel AI system that efficiently and accurately analyzes and detects changes in medical images over time. This research, published on February 20th in the Proceedings of the National Academy of Sciences, demonstrates LILAC's broad potential across various medical applications.
Traditional medical image analysis methods often require extensive customization and preprocessing. For example, with brain MRI data, researchers typically spend considerable time adjusting and correcting images to focus on specific areas, even eliminating variations in angle and size. The LILAC system significantly simplifies this process by automatically performing these complex preprocessing steps, allowing researchers to more easily analyze long time-series image data.
Image Source Note: Image generated by AI, image licensing provider Midjourney
LILAC's flexibility lies in its adaptability to various medical images. The research team trained LILAC using hundreds of microscopic images of in-vitro fertilization embryos, testing its ability to determine temporal order in random image pairs. Results showed LILAC achieved 99% accuracy. In other experiments, the system successfully detected differences in wound healing and changes in the brains of older adults, accurately predicting cognitive scores.
Dr. Hee-Jong Kim, the study's lead designer, stated that LILAC aims to support situations where the research process is not fully understood, especially where significant inter-individual variability exists. This technology is not only applicable to current image data but can also flexibly adapt to future, unknown changes.
Currently, the research team plans to apply LILAC to real-world clinical scenarios, particularly predicting the treatment response of prostate cancer patients through MRI scans. The introduction of this innovative system undoubtedly brings new hope and possibilities to medical image analysis.
