New AI Tool TLPath Predicts Biological Aging by Analyzing Cellular Structures in Routine Biopsy Images

Researchers develop TLPath, an AI model that predicts biological age and telomere length using routine biopsy images, offering a scalable tool for aging research.

By: AXL Media

Published: Mar 17, 2026, 4:19 AM EDT

Source: Information for this report was sourced from News-Medical

Decoding the Architecture of Biological Aging

A groundbreaking computational tool known as TLPath is transforming how researchers measure the biological aging process by analyzing the physical structure of cells. Developed by the Sanford Burnham Prebys Medical Discovery Institute, the model is based on the proven hypothesis that the gradual shortening of telomeres—the protective caps at the ends of DNA—leaves measurable imprints on cellular and tissue morphology. While telomeres naturally whittle down during cell division to protect essential genetic information, their length is a critical predictor of chronological age and the risk of chronic, age-associated diseases. This new approach shifts the focus from expensive molecular assays to the visual data already contained within routine medical biopsies.

Harnessing Big Data for Model Training

The research team utilized a massive dataset from the National Institutes of Health’s Genotype-Tissue Expression Project to train the TLPath model. The project provided over 5,000 high-resolution histopathology slides from 18 different tissue types, donated by nearly 1,000 individuals. This repository allowed scientists to pair visual cellular features with verified laboratory measurements of telomere length. By processing hundreds of terabytes of imaging data, the researchers were able to teach the algorithm to recognize patterns that correlate with genetic degradation across a diverse range of human organs and tissues.

Advanced Computer Vision and Patch Analysis

The TLPath model operates through a sophisticated process of fragmentation and feature extraction. Each biopsy slide is segmented into an average of 1,387 square "patches," which are then scoured for up to 1,024 distinct structural features. Using advanced computer vision, the model identifies higher-order features—many of which are too subtle for the human eye to detect—and assigns them statistical weights. This method allows the AI to calculate an overall score for each slide that accurately predicts the length of the patient’s telomeres, bypassing the need for direct genetic sequencing or chemical buffering tests.