Clinical Decision Support

Analyzing the morphological and molecular characteristics of biopsies and surgically excised tissues (histology) is considered the gold standard for the diagnosis of various diseases and for grading the aggressiveness of those diseases. For many cancers, this information plays an outsized role in stratifying patients for clinical management, and can result in dramatically different treatment paths. For example, patients with low-grade prostate cancer are candidates for active surveillance, whereas patients with higher-grade prostate cancer are candidates for curative therapy (radiation and/or surgery). 

Unfortunately, there are many ambiguities when interpreting heterogeneous 3D tissue morphologies through the interpretation of 2D histology sections, leading to high levels of interobserver variance amongst pathologists.  This can lead to the overtreatment of certain patients with indolent disease, resulting in unnecessary side effects and financial toxicity to both patients and the healthcare system.  Likewise, the under-treatment of patients with aggressive and late-stage disease leads to preventable morbidity and mortality, along with dramatically increased costs of care. 

As a complement to traditional slide-based 2D pathology, there is a growing interest in nondestructive 3D pathology, made possible through recent technological advancements in reversible optical-clearing and fluorescence labeling protocols, high-throughput 3D microscopy, and big-data analysis methods (including AI). 

At the heart of this movement is a belief that nondestructive 3D pathology can improve concordance and accuracy (diagnosis, prognosis, and predicting treatment response) in the analysis of tissue specimens, thus resulting in superior patient treatments and outcomes. 

Technical benefits of 3D pathology over traditional pathology include: (1) improved sampling of large volumes of tissue rather than sparse sampling with thin slide-mounted sections; (2) volumetric imaging of diagnostically relevant structures; and (3) non-destructive imaging, which allows intact tissue specimens (e.g. core-needle biopsies) to be made fully available for downstream assays.

For AI analyses, we are developing both deep-learning classifiers based on sub-visual 3D features and traditional machine classifiers based on intuitive “hand-crafted” 3D features. The first strategy is powerful, and would likely be adopted by oncologists if well-validated in the clinic, but is less interpretable to pathologists and biologists who desire explainable insights. The second approach uses image-processing methods (traditional or deep-learning-based) to segment 3D tissue structures/primitives (e.g. cells, nuclei, glands, collagen, etc.), from which intuitive features (e.g. density, tortuosity, cell-to-cell interactions, etc.) can be extracted.  These histomorphometric features can then be used to train a classifier to yield explainable insights for pathologists and biologists.  

We are also developing AI-triaged 3D pathology pipelines to identify high-risk 2D cross sections within 3D pathology datasets for prioritized pathologist review.  This provides a potential low-risk route towards rapid clinical adoption that continues to rely upon pathologists for diagnostic determinations.  We are showing that such methods can improve the diagnostic accuracy of pathologists compared to standard slide-based 2D histopathology while minimizing pathologist workloads.