Pathologists diagnose and grade prostate cancer using thin 2-dimensional (2D) histologic sections, but these 3 to 5 micron sections are too thin to visualize complete glandular networks and 3-dimensional (3D) spatial relationships of adenocarcinomas. We hypothesized that understanding volumetric glandular organization would reveal architectural features associated with prostate cancer progression and biochemical recurrence (BCR). We analyzed 2 archived prostatectomy cohorts using different sampling methods: simulated 1 mm core-needle biopsies from the University of Washington and 3 × 1 mm-punch biopsies from the University of Pennsylvania. We used open-top light-sheet microscopy to visualize intact tissue networks and developed GlaSkeN, a computational pathology framework to quantify 3D prostatic gland architecture. GlaSkeN used deep learning to segment glandular structures from 3D images and then constructed skeleton-based representations to extract volumetric features, including branch length, branching angles, torsion, and curvature. We analyzed associations between architectural features and 5-year BCR-free survival using 6-fold cross-validated Cox regression. GlaSkeN identified 3D architectural features significantly associated with BCR in both cohorts: the University of Washington (hazard rati [HR], 5.18; 95% CI, 1.18-22.68; C-index = 0.68; P = .019) and the University of Pennsylvania (HR, 2.04; 95% CI, 1.14-3.65; C-index = 0.62; P < .05). In multivariable analysis, GlaSkeN remained prognostic after controlling for clinicopathological variables (HR, 2.30; 95% CI, 1.13-4.7; P = .021). Limitations include different sampling methods between cohorts and limited sample sizes. This 3D analysis captured glandular organization, spatial connectivity, and branching patterns unassessable in 2D cross-sections. GlaSkeN identified glandular architecture features associated with BCR independent of standard clinical variables, suggesting 3D architecture could provide additional prognostic information to complement current histopathological grading. Validation in larger independent cohorts is warranted.

Cellular interactions underlie fundamental biological processes but are not fully represented in conventional 2D histology images. While 3D pathology allows for more-accurate construction of cell-level graphs, machine-learning models are computationally unwieldy and prone to overfitting, especially when dealing with small cohorts. Here, we introduce SCALE3D, a SuperCell graph Analysis framework for LargE 3D pathology datasets. In SCALE3D, spatially adjacent and morphologically similar cells are grouped into functional "supercells." Supercell subtypes are defined via morphology-based clustering and 3D graphs connecting these supercells are used to model their interactions. Validation was performed with 76 radical prostatectomy specimens from patients with known 5-year biochemical recurrence (BCR) outcomes. SCALE3D-derived features achieve higher performance for BCR prediction than established 3D nuclear and glandular morphological features. Combining these complementary features further improves prediction performance. Compared to individual cell-level 3D graphs, SCALE3D maintains comparable prognostic performance with improved noise tolerance while reducing computational times by up to 1,000-fold.

Traditional micro- and macrodissection techniques enable the extraction of localized regions in thin tissue sections for molecular analysis. Despite the growing use of three-dimensional (3D) microscopy, analogous methods for volumetric microdissection are lacking. Here we have developed a 3D microdissection method based on computer numerical controlled milling integrated with open-top light-sheet microscopy. We demonstrate the ability to study tumor evolution along convoluted 3D branching architectures, which is inaccessible to two-dimensional methods.

Current methods for studying tumor phylogeny primarily rely on manual macrodissection or laser capture microdissection of tumor regions identified in thin 2D histology sections. With the emergence of non-destructive 3D pathology workflows, the study of 3D spatial tumor phylogeny is proposed, whereby distinct tumor regions may be macrodissected based on 3D pathology datasets for downstream DNA sequencing and clonal reconstruction. The tumor phylogenetic tree can then be mapped into 3D space using the spatial locations of different subclones, providing an improved understanding of the spatial spread of tumors along their evolutionary paths. To implement the concept of 3D spatial tumor phylogeny, we are developing a 3D pathology-guided macrodissection (3DPM) system as an alternative to current techniques that are limited to thin 2D tissue sections. In our workflow, tissues stained with H&E-analog fluorescence dyes are first imaged using an open-top light sheet (OTLS) microscope. Based on 3D pathology datasets digitally false colored to mimic the appearance of H&E staining, regions of interest with distinct tumor morphologies are annotated by pathologists. The tissue specimen along with the digital annotations are then transferred to a co-registered 3D macrodissection system, which is based on a portable desktop five-axis CNC milling machine. For each volume of interest, the unwanted background tissue is first milled away from the top, followed by detachment of the volume through angled cuts. The toolpath is optimized to maximize the purity and volume of the dissected tissue. Large FFPE tissue slices from archived prostatectomies (centimeter-scale lateral dimensions and millimeter-scale axial dimensions) are being used for proof-of-concept demonstrations. First, we aim to show that the 3D macrodissection method can achieve high tumor purity comparable to laser capture microdissection, despite the increased challenge of dissecting 3D contours vs. 2D tissue regions. Second, we anticipate that more distinct subclones can be identified in each tumor region due to improved sampling of volumetric tissue regions with distinct morphologies (e.g. different Gleason patterns and sub-variants) in comparison to 2D thin-section dissection methods. Finally, given that prostate cancer is known to be multifocal and exhibits complex clonal expansion within the primary tumor, we aim to elucidate whether these multiclonal expansions are spatially separated, intermixed, or vary between cases using reconstructed 3D spatial phylogenetic trees. In summary, 3DPM aims to deepen our understanding of prostate cancer evolution and to enable many new directions in tumor phylogenetics research. Huai-Ching Hsieh, Qinghua Han, Gan Gao, David Brenes, Elena Baraznenok, Robert Serafin, Kevin W. Bishop, Eric Q. Konnick, Colin Pritchard, Sandy Figiel, Freddie C. Hamdy, Ian G. Mills, Lawrence D. True, Michael C. Haffner, Srinivasa R. Rao, Dan J. Woodcock, Jonathan T.C. Liu. Volumetric analysis of spatial tumor phylogeny enabled by 3D pathology-guided macrodissection (3DPM) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2680.

Prostate cancer (PCa) treatment decisions rely heavily on the examination of 2D histology sections (i.e. Gleason grading). However, the limited sampling of specimens afforded by 2D histopathology, and the ambiguities of viewing cross-sections of complex structures such as prostate glands, can cause high interpathologist variability and nonoptimal treatment decisions, especially for low- to intermediate-grade PCa. Our lab has previously shown that computational analysis of 3D histomorphometric features, such as those derived from gland and nuclear segmentation, can improve PCa risk assessment compared to analogous 2D features. Here, we expand on these findings by exploring the prognostic value of 3D features from nerves. These structures are critical as prostate cancer cells follow them to migrate and metastasize, i.e. perineural invasion (PNI), which is correlated with worse patient outcomes. We therefore aim to show the analysis of 3D features related to PNI can provide additional prognostic value. To analyze 3D nerve structures, we utilized a 3D deep learning-based segmentation model, nnU-Net, trained on 3D data from punch biopsies extracted from archived formalin-fixed paraffin-embedded (FFPE) prostatectomy specimens and imaged with a 4th-generation open-top light-sheet (OTLS) microscope. To train a segmentation model without requiring tedious manual annotations, prostate specimens were triple labeled with a fluorescent analog of H&E (nuclear and cytoplasmic stains) plus an antibody targeting PGP9.5, which labels nerves. The H&E-analog channels serve as inputs for the nnU-Net model. To train the model, ground-truth 3D segmentation masks were generated from the PGP9.5 immunofluorescence channel. The trained model allows us to generate 3D segmentation masks of nerves directly from specimens labeled with small-molecule (quickly diffusing) fluorescent analog of H&E, which is fast and inexpensive compared with thick-tissue immunolabeling. The trained 3D nerve segmentation model achieved an average Dice score of 0.86 on held-out validation datasets of tri-labeled specimens (n=10). The segmentation performance was also evaluated on 2D regions (n=53) extracted from 3D pathology datasets of 8 prostatectomy specimens, yielding an average Dice score of 0.64 compared against annotations of nerves generated by a board-certified pathologist. The trained model has been applied to 3D pathology datasets of 120 archived prostatectomy specimens from patients with known biochemical recurrence (BCR) outcomes. We are extracting 3D histomorphometric features (i.e. spatial features) from the resulting nerve segmentations and are integrating them with 3D features from previously developed segmentations of glands, nuclei, and/or cancer-enriched tissue regions. We aim to show that 3D features related to PNI are superior to analogous 2D features for PCs prognostication. Sarah S. Chow, Rui Wang, Yujie Zhao, Robert Serafin, Elena Baraznenok, Lydia Lan, Xavier Farre, Kevin Bishop, Gan Gao, Lawrence D. True, Anant Madabhushi, Jonathan T. Liu. Prostate cancer risk stratification based on 3D histomorphometric features related to perineural invasion (PNI) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2436.

We present an annotation free deep-learning-assisted segmentation pipeline to automatically identify healthy and malignant glands in 3D microscopy images of prostate biopsies stained with fluorescent analogs of H&E.

For AI-analysis of 3D pathology datasets, delineating benign and cancerous tissue regions is often a critical first step. For prostate cancer (PCa) this can be challenging due to the complex intermixing of benign and cancerous glands in 3D. Automated detection of cancerous glands could improve the efficiency and accuracy of downstream computational tasks, such as risk stratification using 3D histomorphometric features. For example, we previously developed machine classifiers based on 3D glandular and nuclear morphologies in PCa biopsies but relied on pathologists to manually identify cancer-enriched volumes for analysis, which was tedious and could introduce bias. Here, we present an approach to automatically identify cancer-enriched regions in 3D pathology datasets of prostate tissue. The tissues are labeled with inexpensive small-molecule fluorescence analogs of H&E staining and are imaged with open-top light-sheet (OTLS) microscopy. Generative adversarial networks were used to train separate models to convert our H&E-analog datasets into synthetic 3D immunofluorescence datasets of two distinct biomarkers: a low-molecular weight cytokeratin (CK8) expressed by luminal epithelial cells found in all prostate glands, and a high-molecular weight cytokeratin (CK5) expressed by basal epithelial cells that are only found in benign prostate glands. Each 3D image-translation model was trained on OTLS microscopy images of prostate tissues that were tri-labeled with fluorescent analogs of H&E plus the antibody of interest (i.e. CK5 or CK8). The models achieved high accuracy for synthetic immunolabeling in held-out validation datasets (Dice scores of 0.72 and 0.83 respectively). By training these models to predict the expression of each CK target directly from H&E-analog datasets, we can avoid the time and cost of immunolabeling large intact tissues. The 3D image-translation models were applied to 3D pathology datasets of 3-mm diameter punch biopsies extracted from archived prostatectomies from 100 PCa patients with known biochemical recurrence outcomes. Using the predicted expression of CK5 and CK8, we developed a simple algorithm to generate a spatial heatmap of cancer-enriched regions throughout each biopsy. For validation, we compared these 3D heatmaps against annotations from 3 GU pathologists of 40 regions of interest, yielding an average Dice score of 0.82. We plan to show improved prognostication with classifiers trained on 3D histomorphometric features derived from cancer-enriched regions (identified by our method) vs. all regions of a prostate specimen. Robert B. Serafin, Jennifer Salguero-Lopez, Rui Wang, Sarah Chow, Elena Baraznenok, Lydia Lan, Kevin W. Bishop, Michelle Downes, Xavier Farre, Lawrence D. True, Anant Madabhushi, Jonathan T. Liu. Automatic detection of prostate cancer in 3D pathology datasets based on synthetic immunolabeling of cytokeratins [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2435.

Recent advances in 3D pathology offer the ability to image orders-of-magnitude more tissue than conventional pathology while providing a volumetric context that is lacking with 2D tissue sections, all without requiring destructive tissue sectioning. Generating high-quality 3D pathology datasets on a consistent basis is non-trivial, requiring careful attention to many details regarding tissue preparation, imaging, and data/image processing in an iterative process. Here we provide an end-to-end protocol covering all aspects of a 3D pathology workflow (using light-sheet microscopy as an illustrative imaging platform) with sufficient detail to perform well-controlled preclinical and clinical studies. While 3D pathology is compatible with diverse staining protocols and computationally generated color palettes for visual analysis, this protocol will focus on a fluorescent analog of hematoxylin and eosin (H&E), which remains the most common stain for gold-standard diagnostic determinations. We present our guidelines for a broad range of end-users (e.g., biologists, clinical researchers, and engineers) in a simple tutorial format.

Consistently generating high-quality datasets across large sample cohorts is necessary for clinical translation of 3D pathology. We present an end-to-end workflow for non-destructive 3D pathology with an emphasis on quality control.

We implement a 3D segmentation workflow on volumetric prostate cancer datasets that involves training a deep learning model to generate synthetic immunofluorescence images highlighting vessels or nerves. The 3D analysis of prostate cancer cells in relation to vessels and nerves is being explored for patient risk assessment.

Open-top light-sheet (OTLS) microscopy offers rapid 3D imaging of large optically cleared specimens. This enables nondestructive 3D pathology, which provides key advantages over conventional slide-based histology including comprehensive sampling without tissue sectioning/destruction and visualization of diagnostically important 3D structures. With 3D pathology, clinical specimens are often labeled with small-molecule stains that broadly target nucleic acids and proteins, mimicking conventional hematoxylin and eosin (H&E) dyes. Tight optical sectioning helps to minimize out-of-focus fluorescence for high-contrast imaging in these densely labeled tissues but has been challenging to achieve in OTLS systems due to trade-offs between optical sectioning and field of view. Here we present an OTLS microscope with voice-coil-based axial sweeping to circumvent this trade-off, achieving 2 µm axial resolution over a 750 × 375 µm field of view. We implement our design in a non-orthogonal dual-objective (NODO) architecture, which enables a 10-mm working distance with minimal sensitivity to refractive index mismatches, for high-contrast 3D imaging of clinical specimens.

A deep learning-model based on the nnU-Net framework was trained for 3D prostate gland segmentation. Compared to our previous ITAS3D pipeline, nnU-Net operation is simpler, faster, and can maintain good accuracy with lower-resolution inputs.

Human tissue, which is inherently three-dimensional (3D), is traditionally examined through standard-of-care histopathology as limited two-dimensional (2D) cross-sections that can insufficiently represent the tissue due to sampling bias. To holistically characterize histomorphology, 3D imaging modalities have been developed, but clinical translation is hampered by complex manual evaluation and lack of computational platforms to distill clinical insights from large, high-resolution datasets. We present TriPath, a deep-learning platform for processing tissue volumes and efficiently predicting clinical outcomes based on 3D morphological features. Recurrence risk-stratification models were trained on prostate cancer specimens imaged with open-top light-sheet microscopy or microcomputed tomography. By comprehensively capturing 3D morphologies, 3D volume-based prognostication achieves superior performance to traditional 2D slice-based approaches, including clinical/histopathological baselines from six certified genitourinary pathologists. Incorporating greater tissue volume improves prognostic performance and mitigates risk prediction variability from sampling bias, further emphasizing the value of capturing larger extents of heterogeneous morphology.

Significance:In recent years, we and others have developed non-destructive methods to obtain three-dimensional (3D) pathology datasets of clinical biopsies and surgical specimens. For prostate cancer risk stratification (prognostication), standard-of-care Gleason grading is based on examining the morphology of prostate glands in thin 2D sections. This motivates us to perform 3D segmentation of prostate glands in our 3D pathology datasets for the purposes of computational analysis of 3D glandular features that could offer improved prognostic performance. Aim:To facilitate prostate cancer risk assessment, we developed a computationally efficient and accurate deep learning model for 3D gland segmentation based on open-top light-sheet microscopy datasets of human prostate biopsies stained with a fluorescent analog of hematoxylin and eosin (H&E). Approach:For 3D gland segmentation based on our H&E-analog 3D pathology datasets, we previously developed a hybrid deep learning and computer vision-based pipeline, called image translation-assisted segmentation in 3D (ITAS3D), which required a complex two-stage procedure and tedious manual optimization of parameters. To simplify this procedure, we use the 3D gland-segmentation masks previously generated by ITAS3D as training datasets for a direct end-to-end deep learning-based segmentation model, nnU-Net. The inputs to this model are 3D pathology datasets of prostate biopsies rapidly stained with an inexpensive fluorescent analog of H&E and the outputs are 3D semantic segmentation masks of the gland epithelium, gland lumen, and surrounding stromal compartments within the tissue. Results:nnU-Net demonstrates remarkable accuracy in 3D gland segmentations even with limited training data. Moreover, compared with the previous ITAS3D pipeline, nnU-Net operation is simpler and faster, and it can maintain good accuracy even with lower-resolution inputs. Conclusions:Our trained DL-based 3D segmentation model will facilitate future studies to demonstrate the value of computational 3D pathology for guiding critical treatment decisions for patients with prostate cancer.

High-quality optical sectioning is required for volumetric imaging of densely labeled clinical tissues. We describe an optimized open-top light-sheet (OTLS) microscope with axially swept illumination to provide improved optical sectioning over large fields of view.

Supplementary Data from Prostate Cancer Risk Stratification via Nondestructive 3D Pathology with Deep Learning–Assisted Gland Analysis