New AI Speeds Up Cancer Cell Detection by 5x

When doctors examine tissue samples for cancer, they look for cells that are dividing rapidly—a hallmark of tumors. But scanning entire slides at high resolution is time-consuming and computationally intensive. Now, researchers have created an AI tool called RapidMitDet that dramatically speeds up this process by using a smart two-step approach: first, it quickly scans the slide at lower resolution to spot potential areas of cell division; then, it zooms in on those areas to confirm. This method is up to 5.6 times faster than the best existing techniques, according to the study.

The key innovation is a technique called 'magnification-aligned knowledge distillation,' which helps the AI transfer what it learns from high-resolution images to lower-resolution ones, reducing errors. Additionally, an 'adaptive triplet learning' method improves the AI's ability to tell apart dividing and non-dividing cells. In tests, the AI's results closely matched those of human pathologists, with high agreement scores (0.879 at 40x magnification and 0.822 at 20x). This means it could soon assist in real clinical settings, helping pathologists work more efficiently.

For patients, this could translate to faster diagnoses and more timely treatment decisions. For pathologists, it means less time spent on tedious scanning and more time on complex cases. The AI is also lightweight, so it can run on standard equipment without needing expensive hardware upgrades. The researchers have made their code publicly available, encouraging further development and validation. While more testing is needed before widespread adoption, this is a promising step toward AI-assisted cancer diagnosis that is both accurate and practical.