AI and Cancer Research ── Prostate ── 2026-10-12
Prostate cancer: treatment, trials and AI
- Abstract
PURPOSE: To systematically review MRI-derived radiomic and artificial intelligence models predicting outcomes in men on or eligible for active surveillance (AS) for prostate cancer, and whether reported endpoints permit quantitative synthesis. METHOD: Following PRISMA 2020, we searched PubMed, Scopus, Embase, and Web of Science (2000-September 2026). Two reviewers independently screened, extracted data, and assessed risk of bias (QUADAS-2, PROBAST) and methodological quality (METRICS). Three subgroups were defined: AS selection at diagnosis, baseline detection of clinically significant cancer, and progression during follow-up. Pooling was pre-specified for subgroups with at least four independent cohorts; endpoint equivalence was assessed. RESULTS: Seventeen studies across 12 unique cohorts were included. Seven publications derived from two cohorts; 12 independent cohort-level estimates (n = 2703) remained after overlap resolution. Nine publications (four independent cohorts) addressed progression (AUC 0.70 to 0.95), but progression was defined in five non-equivalent ways, from any grade-group increase to composite core-count criteria, precluding quantitative synthesis. Risk of bias was high or unclear in 16 of 17 studies; the median METRICS score was 32.3 % (range 18.9-52.3 %). No study reported external validation or calibration for progression. CONCLUSIONS: Progression endpoints were heterogeneous and several were of limited clinical relevance. A core outcome set for AS (at least ISUP grade group ≥3 and/or unequivocal radiological progression), followed by multi-centre external validation with a head-to-head comparison against the PRECISE v2 (Prostate Cancer Radiological Estimation of Change in Sequential Evaluation) score, is the priority before clinical use can be considered.
Journal IF-equivalent: 3.6 (OpenAlex 2-year mean citedness, value as of 2026-10-11, retrieved 2026-10-12; not the official Clarivate IF)Reference: Kyle ET, Sweeney KJ, McCabe DH, O’Sullivan NJ, Corr A, Sheehy N, et al. MRI-derived radiomics for prostate cancer active surveillance: a systematic review. European Journal of Radiology. 2026 Dec;205:113291. doi:10.1016/j.ejrad.2026.113291.Checked: Abstract only - Abstract
Foundation models are trained on massive amounts of data to capture complex patterns in images. Subsequently, a wide range of downstream tasks can be adopted with minimal computational resources. We have developed HistoEncoder, a foundation model for prostate cancer digital pathology by pre-training on 48 million prostate tissue tile images. HistoEncoder allows automated extraction of histological features highly predictive of Gleason patterns achieving comparable performance with substantially larger pan-cancer foundation models while being much more efficient. By fine-tuning the model with a small amount of data and computational resources, we describe two clinical use cases for HistoEncoder. First, HistoEncoder can be used to automatically annotate large-scale datasets with high accuracy. Second, we show that HistoEncoder-derived histology clusters contain prognostic information of a similar magnitude to Gleason grading in internal cross-validation. Lightweight foundation models such as HistoEncoder allow organizations to build effective clinical software tools without the need for extensive datasets and heavy computing.
Journal IF-equivalent: 3.3 (OpenAlex 2-year mean citedness, value as of 2026-10-11, retrieved 2026-10-12; not the official Clarivate IF)Reference: Pohjonen J, Batouche O, Kantola J, Rannikko A, Sandeman K, Erickson A, et al. HistoEncoder: A digital pathology foundation model for prostate cancer. Journal of Pathology Informatics. 2026 Nov;23:100715. doi:10.1016/j.jpi.2026.100715.Checked: Abstract only