Document resource
Background Analysis of tissue architecture in 3D is critical for understanding disease biology, particularly in immuno-oncology and pathologically similar human ovarian cancer subtypes such as Mesonephric Adenocarcinoma (MA), Mesonephric-like Adenocarcinoma (MLA), Endometrioid Carcinoma (EC). However, traditional tools often struggle with capturing and quantifying volumetric, large scale datasets. We present AlpenInsight 3D, a scalable, explainable pipeline designed for discovery-driven spatial tissue analysis using 3D volumetric images captured with light-sheet microscopy of tissues stained with nuclear and eosin structural biomarkers.Methods Human ovarian tumor FFPE tissue biopsies (n=9) were deparaffinized and processed before being stained with a nuclear dye, TO-PRO-3, and Eosin, a cytoplasmic/structural dye. Samples were optically cleared using a modified iDISCO+ protocol. Entire samples were imaged at 4 µm/pixel resolution with a 3Di™ hybrid open-top light-sheet microscope. Light-sheet imaging allowed for non-destructive high-throughput 3D imaging of large of the intact biopsies. Next we adapted 3D Local Binary Patterns (LBP), a texture-based computer vision technique, to extract spatial features from the captured volumetric images. These structural features were integrated with intensity-based signals from Eosin and To-PRO-3 channels. The combined feature sets were then subjected to unsupervised clustering to segment tissue into spatially coherent domains, without the need for predefined labels or supervised training data.Results The resulting clusters reveal distinct spatial domains within tissue volumes. This unsupervised segmentation approach has demonstrated robustness in exploratory contexts, successfully identifying biological heterogeneity across three human ovarian cancer subtypes. While all three MLA samples consistently clustered together, forming a distinct and cohesive group, the MA and EC samples divided into three subgroups: one enriched for MA, one enriched for EC, and a third containing one MA and one EC sample, suggesting overlapping or transitional expression patterns. This indicates that while MLA exhibits a stable signature, MA and EC may share emerging or intermediate profiles.Conclusions AlpenInsight 3D enables automated exploration of whole-biopsy tissue organization in 3D. By combining unsupervised learning with explainable, biologically grounded features, this Methods provides a powerful tool for hypothesis generation and discovery in immuno-oncology and spatial biology. It is particularly well-suited for scalable analysis of volumetric datasets where conventional approaches may fail.Acknowledgements This research was performed in the Flow Cytometry & Cellular Imaging Core Facility, which is supported in part by the National Institutes of Health through M. D. Anderson’s Cancer Center Support Grant P30 CA016672, the NCI’s Research Specialist 1 R50 CA243707-01A1, and a Shared Instrumentation Award from the Cancer Prevention Research Institution of Texas (CPRIT), RP121010.Ethics Approval WSIs of H&E-stained treatment-naïve tumor sections and clinicopathological characteristics from the MDACC dataset were obtained from the ovarian cancer repository of the Department of Gynecologic Oncology and Reproductive Medicine under protocols approved by the University of Texas MD Anderson’s Institutional Review Board. Written informed consent from the patients were obtained by front desk personnel, and the studies were conducted in accordance with recognized ethical guidelines