A new AI-based system for analyzing images that are taken over time can accurately detect changes and predict the results, according to a study led by researchers by Weill Cornell Medicine, Cornell’s Ithaca campus and Cornell Tech. The sensitivity and flexibility of the system can make it useful for a wide range of medical and scientific applications.
The new system, called Lilac (on leather-based inference of longitudinal image changes), is based on an AI approach called Machine Learning. In the study, which will be published on 20 February in the procedure of the National Academy of Sciences, the researchers developed the system and demonstrated it at various time series of images, called “longitudinal” image-covering developing IVF-Embryos, healing tissue after wounds and aging brains. The researchers showed that Lila has a broad capacity to even identify very subtle differences between images taken at different times, and to predict related outcome measures such as cognitive scores of brain scans.
“This New Tool Will Allow Us to DECTECT AND Quantify Clinical Relevant Changes About Time In Ways That Weren’t Possible Before, And Its Flexibility Means That It Can Be Applied Off-The-Shelf To Virtual Senior Authordal.” Said Dataset, ” Mert Sabuncu, Vice Chair of Research and A Professor of Electrical Engineering at Radiology at Weill Cornell Medicine and Professor in the School of Electrical and Computer Engineering at Cornell University Campus and Cornell Tech.
The first author of the study is Dr. Heejong Kim, an instructor of artificial intelligence in radiology at Weill Cornell Medicine and member of the Sabuncu Laboratory.
Traditional methods for analyzing longitudinal image data sets usually require extensive adjustment and pre -processing. Researchers who study the brain can, for example, take raw brains from the brain and process the image data in order to concentrate on only one brain area, which also correct for different areas of display, dimensioning differences and other artifacts in the data all before they perform the main analysis.
The researchers have designed Lila to work much more flexibly, effectively automatically implement such corrections and find relevant changes.
As a result, Lila can not only be useful in different imaging contexts, but also in situations where you are not sure what change you want to expect.
Dr. Heejong Kim, the most important designer of Lilac, instructor of artificial intelligence in radiology at Weill Cornell Medicine
In one proof-of-concept demonstration, the researchers trained Lila on hundreds of sequences of microscope images that develop in vitro-free embryos, and then tested it against new embryo figuratives. Lilac, for randomized couples of images from a certain series, had to determine which image was taken earlier —— that cannot be performed reliably, unless the image data contain a real “signal” that indicate time-related change. Lilac performed this task with approximately 99% accuracy, the few errors that occur in images with relatively short time intervals.
Lilac was also very accurate when ordering pairs of images of healing tissue from the same sequences, and when detecting differences at group level in healing rates between untreated tissue and tissue that received experimental treatment.
Lilac also predicted the time intervals between MRI images of the brain of healthy older adults, as well as individual cognitive scores of MRIs of patients with mild cognitive impairment in both cases with far fewer errors compared to basic methods.
In all these cases, the researchers showed that Lila can be easily adjusted to emphasize the image characteristics that are most relevant to detecting changes in individuals or differences between groups that can offer new clinical and even scientific insights.
“We expect that this tool will be useful, especially in cases where we have no knowledge about the process that is being studied, and where there is a lot of variability between individuals,” said Dr. Sabuncu.
The researchers are now planning to prove Lila in a Real-World setting to predict treatment reactions of MRI scans of prostate cancer patients.
Source:
Journal Reference:
Kim, H., et Alt Alto. (2025). On leather-based inference of longitudinal image changes: applications in embryo development, wound healing and aging brain. Proceedings of the National Academy of Sciences. doi.org/10.1073/pnas.2411492122.