AI Promises Faster, Scalable Histology for Respiratory Research

Histological analysis—looking at stained tissue under the microscope—is a central tool in preclinical respiratory disease research. It helps scientists see disease patterns, judge whether an experimental treatment is working, and explore disease mechanisms. But the traditional way of doing this relies on people manually scoring slides, a process that can be subjective, slow, and hard to scale because different observers can disagree and throughput is low. To understand how artificial intelligence (AI) is being used to change that picture, researchers led by Manoj M. Lalu conducted a scoping review following Joanna Briggs Institute guidelines. They searched MEDLINE and Embase from their start until January 2025 for preclinical studies that used AI to analyze histology in respiratory disease models. Screening, full-text review, and data extraction were done in duplicate to keep the process rigorous. From 6,271 studies screened, 29 met the review’s inclusion criteria. The review set out to map where AI has been applied, what types of models and stains were used, what AI tasks were performed, and how well those AI tools were validated and reported.

The studies in the review mostly used murine models (76%) and focused on a few lung conditions: lung cancer (28% of included studies), pulmonary fibrosis (24%), and tuberculosis (17%). Hematoxylin and eosin was the most common stain (48%), while other studies used stains targeting collagen or immune markers. AI work fell into three main tasks: image classification (20 studies), segmentation (10 studies), and object detection (4 studies). Deep learning approaches predominated, with convolutional neural networks used in 69% of studies. Preprocessing steps such as stain normalization were commonly reported, but how images were annotated and how models were trained was described inconsistently across papers. Reported performance was often high—seven studies reported accuracy of 90% or greater—but validation approaches varied and no studies performed external validation. Many groups relied on “black box” models and applied few explainability techniques. Measures to support reproducibility, like sharing datasets or code, were rarely reported. The review itself is registered on the Open Science Framework https://doi.org/10.17605/OSF.IO/NM94E.

Taken together, the review shows that AI tools are already being used to analyze histology in preclinical respiratory research and that they have the potential to make the work faster and more scalable. At the same time, the review highlights important gaps that must be filled before these tools can be widely trusted and adopted. Key shortcomings include inconsistent reporting of annotation and training practices, a lack of external validation to show models work beyond the original data, the widespread use of “black box” methods with few attempts at explainability, and limited sharing of code and datasets to support reproducibility. The authors conclude that by addressing these gaps—improving validation, increasing transparency, and developing standards for reporting and data sharing—the field can better harness AI to deliver robust, efficient, and scalable histology workflows in preclinical respiratory research.