AI enables much faster pathology for life-or-death interventions

After training deep neural networks on around 4,000 slide images from around 40 biopsied kidney patients, UCLA engineers have virtually re-stained tissue images for speedier high-accuracy diagnostics than a human histotechnologist could support.  

Turnaround time was especially impressive, as the AI completed in seconds a task that typically requires hours if not days—and because minutes matter in critical lab-reliant procedures like kidney transplant rejections.

What’s more, in clinical practice the high efficiency would translate to substantial savings.

The work is described in a study published by Nature Communications.

Senior author Aydogan Ozcan, PhD, and colleagues concentrated on virtualizing specialized stains from image data produced by the most common stain, hematoxylin and eosin (H&E), because H&E is used in most tissue-staining procedures.

They used label-free autofluorescence tissue images to create perfectly registered training images.

The process “simultaneously generated both the H&E and special stain images with a nanoscopic match in the local coordinates of each virtually stained image pair of our training dataset,” the authors explain.

The resulting technique is highly generalizable, they state, to other “stain-to-stain” image transformations.

“For example, transformations from special stains to H&E or from immunofluorescence to H&E or special stains could be performed using the presented method,” Ozcan and co-authors write. “Our approach allows pathologists to visualize different tissue constituents without waiting for additional slides to be stained with special stains, and we demonstrated it to be effective for the clinical diagnosis of multiple renal diseases.”