PRECISE: Benchmarking digital pathology with expert-annotated contiguous IHC-H&E serial prostate sections
The paper introduces PRECISE, the first publicly available dataset of expert-annotated, spatially harmonized H&E and IHC serial prostate biopsy sections, designed to serve as a robust multimodal benchmark for advancing AI-assisted diagnosis through pixel-level semantic segmentation of the full morphological spectrum of prostate pathology.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Imagine you are a detective trying to solve a mystery inside a tiny, invisible city. This city is made of cells, and sometimes, a few of them turn into troublemakers called cancer. To find them, doctors use a special kind of microscope that takes giant, high-definition photos of these cell cities. These photos are called "Whole Slide Images." Usually, the detective uses a standard blue-and-pink paint (called H&E) to color the cells. It's like looking at a black-and-white map; you can see the shapes, but sometimes the troublemakers look so much like the good guys that it's hard to tell them apart. When the detective gets confused, they ask for a second opinion using a special "highlighter" paint (called IHC) that glows on specific cells to prove who is who. This two-step process—looking at the map, then checking the highlighter—is how doctors currently solve the hardest cases. But teaching computers to do this detective work has been tough because most computer training sets only show the blue-and-pink map, leaving the computer guessing when the highlighter is needed.
This is where a new project called PRECISE steps in. Think of PRECISE as the first-ever "bilingual" training manual for computer detectives. Instead of just giving the computer one picture, the researchers created a dataset where every single blue-and-pink map is paired perfectly with its matching highlighter map. They took 37 tiny samples of prostate tissue from 25 patients and created 27 sets of these paired images. What makes this special is that two expert pathologists (the human detectives) didn't just look at them; they drew detailed maps on both the blue-and-pink and the highlighter versions, marking exactly where the cancer, the good cells, and the tricky "almost-cancer" cells are. They even used the highlighter paint as a "truth serum" to make sure their maps were 100% accurate. The result is a massive library of 24,387 labeled regions covering seven different types of tissue, including rare and confusing ones that other computer datasets often ignore.
The paper doesn't claim that computers can now cure cancer or that this dataset solves every problem. Instead, it suggests that this new resource is a powerful tool to help researchers build better AI. The authors measured the data and found that while computers can sometimes tell cancer from good cells just by color, there is still a lot of overlap that makes it hard to be sure without the expert's eye. By providing these paired, expert-annotated images, the team hopes to help AI learn to spot the subtle differences that humans see, especially in those tricky cases where the standard map isn't enough. They are careful to note that because the data comes from just one hospital and a specific group of patients, the AI might need extra practice before it can work perfectly everywhere else. But for now, PRECISE offers a unique, open invitation for scientists to train their digital detectives on the real, two-step way doctors actually work, potentially leading to faster and more accurate diagnoses in the future.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.