Active surveillance decisions in intermediate-risk prostate cancer depend on identifying tumour features that may make monitoring unsafe. Cribriform growth is one such feature, but biopsy can miss it. Current risk grouping relies on clinical factors and biopsy findings, without routinely using imaging information. A recent analysis published in European Radiology Experimental evaluates whether MRI-based radiomics can help identify patients without cribriform growth who may be suitable for active surveillance rather than active treatment.
Radiomics Adds Imaging to Risk Assessment
The analysis included 127 men with prostate cancer who had undergone MRI and radical prostatectomy at the Netherlands Cancer Institute. The cohort covered lower-risk and selected intermediate-risk categories, including patients with Cambridge Prognostic Group 1, Cambridge Prognostic Group 2 and Cambridge Prognostic Group 3 disease with Gleason grade 2. Radical prostatectomy specimens provided the reference for whether cribriform growth was present anywhere in the prostate.
Must Read: PSMA-PET Refines Prostate Cancer Risk
The MRI-based model used radiomics information from apparent diffusion coefficient maps. Rather than assessing only a visible lesion, the model assessed the whole prostate and generated probability maps showing where cribriform growth might be present. Each patient then received a maximum probability score, reflecting the most suspicious area identified by the model.
The model had moderate standalone performance in identifying cribriform growth. However, the main aim was not simply to test diagnostic accuracy. The central question was whether the model could support patient selection for active surveillance when added to structured clinical scenarios. This made the imaging output relevant to simulated treatment allocation, separating patients who could be monitored from those who would remain directed towards active treatment.
Structured Scenarios Test Surveillance Expansion
Two active surveillance scenarios were simulated. In the first, Cambridge Prognostic Group 1 patients remained eligible for surveillance, while the model was used to assess selected Cambridge Prognostic Group 2 patients. Those without radiomics-predicted cribriform growth entered the surveillance pathway, while those with predicted cribriform growth stayed in the active treatment pathway.
The second scenario widened the same approach to include selected Cambridge Prognostic Group 3 patients with Gleason grade 2. Again, surveillance eligibility depended on the absence of radiomics-predicted cribriform growth. A reference scenario placed only Cambridge Prognostic Group 1 patients on active surveillance, with all other patients directed towards active treatment.
The comparison assessed whether radiomics-guided allocation could reduce overtreatment without producing an unacceptable rise in undertreatment. Overtreatment referred to patients without cribriform growth and without higher-grade disease who would still have received active treatment. Undertreatment referred to patients who would have entered surveillance despite cribriform growth or higher-grade disease on final pathology. Under these assumptions, the model allowed more patients to enter simulated surveillance while maintaining a focus on histological features that made surveillance unsuitable.
Threshold Choice Shapes Clinical Trade-Offs
A probability threshold of 0.60 gave the most favourable balance between overtreatment and undertreatment in both scenarios. In the scenario including Cambridge Prognostic Group 1 and 2 patients, radiomics guidance increased appropriate allocation from 61% to 69%. It also increased the share of patients eligible for surveillance and reduced overtreatment, with only a small rise in undertreatment.
In the broader scenario that also included selected Cambridge Prognostic Group 3 patients with Gleason grade 2, the same threshold again increased appropriate allocation from 61% to 69%. Overtreatment fell, while undertreatment remained at the same low level as the reference scenario. These findings show that a model with moderate standalone accuracy can still improve treatment allocation when used within a defined decision framework.
Raising the threshold above 0.60 reduced overtreatment further but increased undertreatment. Many missed cases involved small cribriform regions below 1.5 mm, which may be difficult for MRI to identify. Some patients without cribriform growth still had higher-grade disease at prostatectomy, which made them unsuitable for surveillance under the stated criteria. The model therefore addresses cribriform growth specifically and does not replace assessment of other adverse pathological features.
MRI-based radiomics may help refine active surveillance decisions in intermediate-risk prostate cancer by supporting non-invasive exclusion of cribriform growth. The model does not offer a standalone decision rule, but it improves simulated allocation when used within structured clinical pathways. At the selected threshold, radiomics guidance reduces overtreatment while preserving low undertreatment. The findings remain preliminary because the work is retrospective, single-centre and based on hypothetical treatment allocation. Larger prospective and multicentre validation is needed before use in routine decision-making.
Source: European Radiology Experimental
Image Credit: iStock
References:
Fernandez Salamanca M, Simões R, Deręgowska-Cylke M et al. (2026) MRI radiomics predicting cribriform growth informs active surveillance decision in intermediate-risk prostate cancer. Eur Radiol Exp; 10, 67.
