Volumetric histopathology of surgical specimens: scalable 3D analysis in breast and prostate cancer - Project Abstract Conventional pathology examines thin, two-dimensional (2-D) sections from large surgical specimens, often sampling well under 1% of the tissue. Important but small or complex three-dimensional (3-D) patterns can be missed or unrecognizable when analyzed this way. For instance, in breast-conserving surgery, standard margin evaluation can miss residual microscopic disease in up to half of cases because of limited 2-D sampling. In prostate cancer, aggressive cribriform gland architecture is strongly linked to metastasis, yet its ≥0.25 mm threshold is far smaller than the typical 3–4 mm section spacing, making detection and reporting inconsistent. To address this fundamental challenge, I propose developing an integrated platform for comprehensive 3-D histology through the combination of tissue clearing, high-throughput two-photon microscopy, and automated 3-D analysis. The research program consists of three specific aims: Aim 1 (K99) will optimize tissue clearing and engineer an automated, high-throughput two-photon microscopy system with adaptive controls to maintain image quality across depth. Aim 2 (K99/R00) will deliver open-source software for acquisition, storage, and viewing of volumetric pathology images, and develop machine-learning methods to segment epithelium, tumor, ducts/glands, nerves, and vessels, producing interpretable 3-D maps. Aim 3 (R00) will evaluate clinical utility in two high-impact settings: (i) lumpectomy margins, testing whether continuous 3-D margin metrics better predict residual disease than conventional 2-D reads; and (ii) radical prostatectomy, assembling whole-gland 3-D ground truth to clarify MRI–pathology correspondence and challenging tumor geometries. This research will deliver a validated, disseminable platform for 3-D surgical pathology; robust, shareable datasets and tools (released openly) that enable multi-site studies; and evidence that volumetric metrics improve risk assessment and treatment planning in breast and prostate cancer. The mentored phase (MIT and partner hospitals) couples formal training in machine learning for volumetric medical data, cancer biology, and best practices for sustainable open-source software with structured clinical immersion and leadership development. This integrated plan equips me to launch an independent R00 program at the interface of biomedical optics, computational pathology, and cancer biology, advancing NCI’s mission to deliver more precise, patient-specific care.
