Researchers at Weill Cornell Medicine have created a new artificial intelligence (AI) tool called LILAC (Longitudinal Implicit Latent Alignment and Clustering) that can analyze image data to detect subtle changes and patterns over time. The tool’s flexibility allows it to be used effectively even when the processes under investigation are not well-understood.
Traditional methods for analyzing longitudinal image datasets often require extensive customization and pre-processing. According to the AI tool’s principal designer, Dr. Heejong Kim, such methods can be inflexible. For example, researchers studying the brain may pre-process raw MRI data to focus on a specific brain area, correcting for viewing angles, size differences and other artifacts. LILAC, on the other hand, is designed to automatically perform these corrections and identify relevant changes, making it adaptable to different imaging contexts.
“This enables LILAC to be useful not just across different imaging contexts but also in situations where you aren’t sure what kind of change to expect,” Dr. Kim said. Dr. Kim is also an instructor of artificial intelligence in radiology at Weill Cornell Medicine and a member of the Sabuncu Laboratory. Another researcher, Mert Sabuncu, added, “We expect this tool to be useful especially in cases where we lack knowledge about the process being studied, and where there is a lot of variability across individuals.”
In a proof-of-concept demonstration, LILAC was trained on hundreds of sequences of microscope images showing in-vitro-fertilized embryos as they develop. When tested with new embryo image sequences, LILAC had to determine which image in a randomized pair was taken earlier. The AI accomplished this with approximately 99% accuracy, with the few errors occurring in pairs with relatively short time intervals. It accurately ordered pairs of images of healing tissue and found group-level differences in healing rates between treated and untreated tissue. In a similar vein, LILAC accurately predicted the time intervals between MRI images of healthy older adults’ brains and individual cognitive scores from MRIs of patients with mild cognitive impairment. In both cases, its predictions were far more accurate than those of baseline methods.
The researchers noted that LILAC can easily highlight the image features most relevant for detecting changes in individuals or differences between groups. This capability could provide new clinical and scientific insights. The researchers now plan to demonstrate LILAC in a real-world setting to predict treatment responses from MRI scans of prostate cancer patients.
