Johnson, Justin M.

Abstract Background: Prostate cancer remains one of the most common malignancies in men, with significant heterogeneity in clinical outcomes. Early and accurate diagnosis and risk stratification is crucial for effective treatment and improved outcomes. Although Gleason score is an established risk stratification tool, it does not fully explain outcome variability. We present a multimodal deep-learning framework that integrates attention-based whole-slide image (WSI) features with a clinical variable (Age) to predict survival in patients with prostate adenocarcinoma. Design: Clinical data and WSI for 500 patients with prostate adenocarcinoma were obtained from TCGA-PRAD. Clinical data were parsed for recurrence or progression (RoP) events and age. RoP was used as the outcome label, and age was used as a patient feature. Diagnostic WSI tumor regions were annotated and tiled at 224×224 pixels using QuPath (Bankhead Sci Rep 2017). Tile embeddings were extracted with UNI2-h (Mahmood Nat Med 2025). An end-to-end machine learning pipeline was developed to aggregate tile embeddings into slide-level embeddings using Clustering-Constrained Attention Multiple Instance Learning (CLAM) and predict RoP (Lu Nat Biomed 2021). A second pipeline used a Gleason score-trained binary cross entropy (BCE) model to predict RoP in the same patients. Three models were evaluated: WSI-only CLAM model, WSI+age CLAM model, and Gleason-only BCE model. All models were trained and evaluated using 5-fold stratified cross-validation. Each fold was repeated 3 times with independent random initialization. AUROC was averaged across all folds and runs per model. Standard deviation was also computed. Model performance was compared using Friedman’s test, with post hoc pairwise comparisons by Wilcoxon signed-rank test. Results: The Gleason BCE model underperformed (AUROC 0.67 ± 0.07). Both WSI-based models outperformed the Gleason BCE model, achieving AUROC 0.76 ± 0.02 (without age), and AUROC 0.76 ± 0.01 (with age). AUROCs were significantly different by Friedman’s test (p = 0.0224) and showed a trend toward significance in post-hoc comparison between WSI and Gleason-based approaches (p = 0.06). Conclusion: AI models using WSI-derived features, with or without basic clinical context, outperformed traditional Gleason score for recurrence risk prediction in prostate cancer. These findings support the integration of digital pathology and AI into routine prognostic assessment for prostate cancer. Future work will focus on including additional patients, clinical variables, and multi-class prognostic labels to expand prognostic prediction beyond RoP. Citation Format: Justin Johnson, Kingsley Ebare. Whole-slide image and clinical feature integration for superior prostate cancer risk stratification [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1441.

Abstract Background: The functional importance of bone marrow adipocytes (BMA) is frequently overlooked even though they are critical components of the BM (bone marrow) microenvironment. BMA have been proposed to play a supportive role within the hematopoietic niche and to directly sustain hematopoietic stem cell (HSC) survival as well as to preferentially support leukemia cell survival and proliferation (Sabbah Cell Commun Signal 2023). Although significant ex-vivo data implicates direct and indirect crosstalk between BMA and neoplastic HSCs (Venti Front Endocrinol 2022), the possible role of BMA in myeloproliferative neoplasms and their dynamics following therapy have yet to be characterized. Adiponectin, an anti- inflammatory adipokine that maintains metabolic homeostasis through regulation of insulin sensitivity, lipid metabolism and pro-inflammatory signaling is low in MPN patients and increases with IFN-α therapy (Avcu Int J Hematol. 2006). IFN-α has also been implicated in inhibiting adipogenesis via JAK-STAT1 signaling (Lee Biochim Biophys Act. 2016). A single institutional study of IFN-α in patients with essential thrombocythemia (ET) and polycythemia vera (PV) confirmed the potential for BM histopathologic responses and a correlation with molecular responses (Masarova Exp Hematol Oncol 2017). The aim of this analysis was to characterize the change in BM cellularity and fat composition with pegylated IFN α-2a (Pegasys; PEG) and the relationship with both hematologic and molecular response from tissue collected in prospective, global clinical trials of PEG in patients with ET/PV. Methods: A total of 24 subjects from the phase 2 MPN-RC 111 (Yacoub Blood 2019) salvage PEG study and 16 subjects from the MPN-RC 112 (Mascarenhas Blood 2022) phase 3 study comparing PEG to HU with available clinical, BM H&E slides, laboratory and molecular data were included in this retrospective analysis of prospective clinical trial data. We assigned a semi-quantitative score for changes in cellularity, fat distribution, and megakaryocyte (MK) histo-topography for changes from baseline, at 12, 24 and 36 months when available. Changes in BMA content and para-trabecular distribution were assessed on whole slide digital images with machine learning models. BMA content was quantified with a QuPath pixel classifier that distinguished adipocytes from other marrow components (Bankhead Scientific Reports 2017). Para-trabecular adipocyte distribution was assessed by leveraging UNI2-h, a self-supervised Vision Transformer (ViT-H/14) model and successor to UNI (Mahmood Nature 2024). UNI2-h was pre-trained on 200 million image tiles from over 300,000 H&E and IHC slides. In this study, UNI2-h was adapted to distinguish between high and low para-trabecular adiposity using image tiles centered on trabeculae. Trabeculae with a surrounding layer of adipocytes were labeled as ‘high’ adiposity, while ‘low’ para-trabecular fat distribution was defined by the direct contact of cellular marrow with trabeculae, lacking surrounding adipocytes. Results: A total of 40 PEG-treated ET (n=18) and PV (n=22) patients were included in this analysis with a median age of 64 years (26-85), 77% JAK2V617F+, 23% CALRmut+, and 40% treatment-naïve. The median JAK2V617F variant allele fraction (VAF) was 26% (range 3-96%). A complete hematologic response (CHR) was obtained with PEG in 58%, 61% attained >20% reduction in driver VAF and 48% attained a >50% reduction in VAF (MR) within 24 months of baseline. We observed an increase in BMA content and notable increase in para-trabecular distribution (confirmed by morphology and ML, independently) in 93% and 53% of PEG treated patients attaining a MR versus only 31% and 6% who did not attain MR, respectively (P Value=0.0004). There was more notable improvement towards normalization of MK histo-topography and orderly erythropoiesis in MR (86.6%) than in non-MR (31.2%) (P value=0.001). Increase in BMA content, BMA para-trabecular distribution, or improvement in MK histo-topography were not associated with achievement of CHR or changes in CALR VAF in this small cohort. Cytokine correlation with BMA dynamics is currently ongoing and will be reported at the meeting. Conclusion: Treatment of MPN patients with IFN-α resulted in reproducible changes in BM cellularity including MK histo-topography, erythropoiesis, BMA number and distribution. Changes in BMA were closely associated with reduction in JAK2V617F VAF following PEG treatment.