BACKGROUND AND PURPOSE: Currently, contrast-enhancing margins on T1WI are used to guide treatment of gliomas, yet tumor invasion beyond the contrast-enhancing region is a known confounding factor. Therefore, this study used postmortem tissue samples aligned with clinically acquired MRIs to quantify the relationship between intensity values and cellularity as well as to develop a radio-pathomic model to predict cellularity using MR imaging data. MATERIALS AND METHODS:This single-institution study used 93 samples collected at postmortem examination from 44 patients with brain cancer. Tissue samples were processed, stained with H&E, and digitized for nuclei segmentation and cell density calculation. Preand postgadolinium contrast T1WI, T2 FLAIR, and ADC images were collected from each patient's final acquisition before death. In-house software was used to align tissue samples to the FLAIR image via manually defined control points. Mixed-effects models were used to assess the relationship between single-image intensity and cellularity for each image. An ensemble learner was trained to predict cellularity using 5 Â 5 voxel tiles from each image, with a two-thirds to one-third train-test split for validation.RESULTS: Single-image analyses found subtle associations between image intensity and cellularity, with a less pronounced relationship in patients with glioblastoma. The radio-pathomic model accurately predicted cellularity in the test set (root mean squared error ¼ 1015 cells/mm 2 ) and identified regions of hypercellularity beyond the contrast-enhancing region. CONCLUSIONS:A radio-pathomic model for cellularity trained with tissue samples acquired at postmortem examination is able to identify regions of hypercellular tumor beyond traditional imaging signatures.

This study was proposed and organized through an NCI working group. Investigators from the central organizing institution and 14 other institutions participated. Data were collected at the central site and distributed to each satellite institution for processing. Fourteen implementations were included in this project from investigators at MCW (Team 2),

No abstract

One in eight men will be affected by prostate cancer (PCa) in their lives. While the current clinical standard prognostic marker for PCa is the Gleason score, it is subject to inter-reviewer variability. This study compares two machine learning methods for discriminating between cancerous regions on digitized histology from 47 PCa patients. Whole-slide images were annotated by a GU fellowship-trained pathologist for each Gleason pattern. High-resolution tiles were extracted from annotated and unlabeled tissue. Patients were separated into a training set of 31 patients (Cohort A, n = 9345 tiles) and a testing cohort of 16 patients (Cohort B, n = 4375 tiles). Tiles from Cohort A were used to train a ResNet model, and glands from these tiles were segmented to calculate pathomic features to train a bagged ensemble model to discriminate tumors as (1) cancer and noncancer, (2) high- and low-grade cancer from noncancer, and (3) all Gleason patterns. The outputs of these models were compared to ground-truth pathologist annotations. The ensemble and ResNet models had overall accuracies of 89% and 88%, respectively, at predicting cancer from noncancer. The ResNet model was additionally able to differentiate Gleason patterns on data from Cohort B while the ensemble model was not. Our results suggest that quantitative pathomic features calculated from PCa histology can distinguish regions of cancer; however, texture features captured by deep learning frameworks better differentiate unique Gleason patterns.

Background: This study identified a clinically significant subset of glioma patients with tumor outside of contrast-enhancement present at autopsy, and subsequently developed a method for detecting non-enhancing tumor using radio-pathomic mapping. We tested the hypothesis that autopsy-based radio-pathomic tumor probability maps would be able to non-invasively identify areas of infiltrative tumor beyond traditional imaging signatures. Methods: A total of 159 tissue samples from 65 subjects were aligned to MRI acquired nearest to death for this study. Demographic and survival characteristics for patients with and without tumor beyond the contrast-enhancing margin were computed. An ensemble algorithm was used to predict pixelwise tumor presence from pathological annotations using segmented cellularity (Cell), extracellular fluid (ECF), and cytoplasm (Cyt) density as input (6 train/3 test subjects). A second level of ensemble algorithms were used to predict voxel-wise Cell, ECF, and Cyt on the full dataset (43 train/22 test subjects) using 5-by-5 voxel tiles from T1, T1+C, FLAIR, and ADC as input. The models were then combined to generate non-invasive whole brain maps of tumor probability. Results: Tumor outside of contrast was identified in 41.5% of patients, who showed worse survival outcomes (HR=3.90, p<0.001). Tumor probability maps reliably tracked non-enhancing tumor in the test set, external data collected pre-surgery, and longitudinal data to identify treatment-related changes and anticipate recurrence. Conclusion: This study developed a multi-stage model for mapping gliomas using autopsy tissue samples as ground truth, which was able to identify regions of tumor beyond traditional imaging signatures.