Exploratory Analysis of Peri-Prostatic Adipose Tissue Radiomics Features and Their Potential to Predict Prostate Cancer Diagnosis: Results From a UK Biobank Study.
Authors
Affiliations (8)
Affiliations (8)
- Applied Radiation Therapy Trinity, Discipline of Radiation Therapy, Trinity College Dublin, Dublin, Ireland.
- Trinity St. James's Cancer Institute, Dublin, Ireland.
- Johnston Centre for Cancer Research, Queen's University Belfast, Belfast, Ireland.
- Department of Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, UK.
- MRC Integrative Epidemiology Unit, Bristol Medical School, University of Bristol, Bristol, UK.
- NIHR Bristol Biomedical Research Centre, University Hospitals Bristol and Weston NHS Foundation Trust and the University of Bristol, Bristol, UK.
- Faculty of Medicine, School of Health Sciences, University of Iceland, Reykjavík, Iceland.
- Cancer Epidemiology Unit, Nuffield Department of Population Health, University of Oxford, Oxford, UK.
Abstract
This paper reports the development and application of an artificial intelligence (AI)-led approach to large volume MRI data curation and image segmentation of periprostatic adipose tissue (PPAT) in prostate cancer (PCa)-free men from the UK Biobank. We report how PPAT was defined, radiomics features of PPAT extracted, and explored their potential to predict a PCa diagnosis. In total, MR images from 31,042 non-PCa men were extracted and made available for autosegmentation. PPAT was defined using tools in 3D Slicer as anterior (Ant PPAT) or posterior (Post PPAT) in relation to Denonvilliers fascia. Dice-Sørensen coefficients (DICE) were calculated to measure the agreement between the observer's manual segmentations used to train a 3D convolutional neural network with U-Net architecture (nnU-Net). Radiomics features were extracted using PyRadiomics and logistic regression analysis was performed on the Ant and Post PPAT features that were deemed significantly different between men without PCa and those who went on to develop PCa. Statistical analysis was performed in SPSS (version 29.0.0). A total of 107 features (14 shape, 18 first-order texture and 75 second-order texture) were extracted from auto-segmentations of Ant PPAT and Post PPAT. A case-control radiomics analysis was performed on 297 men who had an MRI scan when cancer-free (but who subsequently went on to be diagnosed with PCa), age-matched (within 1 year) to participants without a PCa diagnosis during the follow-up period. In Ant PPAT there were statistically significant differences in 31 features compared to 4 in Post PPAT between men without PCa and those with prevalent PCa. Stratifying the data by the prevalent PCa group time to diagnosis had a limited effect on these results. Predictive analysis results were not statistically significant for Ant PPAT or Post PPAT, however are comparable with the reported predictive value of other clinically established biomarkers of PCa. An automated process for data curation and image segmentation was successfully demonstrated. Results suggest that Ant PPAT justifies further investigation as a predictive biomarker for prostate cancer diagnosis.