MEDICAL STUDENT COLUMN: The Use of Magnetic Resonance Imaging-derived Radiomic Models in Prostate Cancer Risk Stratification

By: Linda My Huynh, MSc, University of Nebraska Medical Center, Omaha | Posted on: 09 Mar 2023

In recent years, the advancement of precise medical imaging has facilitated the establishment of radiomics, a computer-based method of extracting and quantifying subvisual imaging characteristics.1 These features (ie, qualities of intensity, texture, shape, or wavelet) can be extracted from a variety of medical images (CT, MRI, or positron emission tomography) using advanced mathematical algorithms, aggregated into predictive models via machine learning, and applied to enhance personalized therapies. In the last decade, several studies have highlighted the enormous potential of radiomics in enhancing care for a variety of diseases. These include, but are not limited to, cancers of the gastrointestinal tract, lung, brain, and (more recently) the genitourinary tract.

In prostate cancer, MRI is a standard clinical tool used for diagnosis, prognosis, and treatment planning. As a key part of the prostate cancer clinical care pathway, MRI represents an opportune point of intersection for the use of radiomics. In this regard, the number of research articles on MRI-derived prostate radiomics has predictably increased since 2017 (see Figure), accounting for 218 original articles in the last 5 years. However, as the technology is still evolving, the exploration of radiomic-based models in prostate cancer has thus far largely been sequestered within the fields of radiation oncology, radiology, and biomedical imaging. Even further, most of these investigations have concentrated on screening or diagnostic uses—ie, the correlation of radiomic models with PI-RADS (Prostate Imaging Reporting and Data System) lesions, in confirming biopsy findings, or in prostate cancer screenings. However, as prostate cancer is highly heterogeneous, the use of radiomics could be further extended to enable prediction beyond initial diagnosis and toward risk stratification, prognostication, and prediction of therapy response.

Figure. Number of prostate cancer radiomics publications from 2017 to 2022.

Of the 218 articles published on MRI-derived prostate radiomics in the last 5 years, 42 (19.3%) have utilized MRI-derived radiomics specifically for prostate cancer risk stratification and prognostication. Prediction of Gleason grade group and adverse pathologies, including seminal vesicle invasion, extraprostatic extension, and lymph node involvement, were primary endpoints in 21 (50%) and 11 (26.2%) published articles from 2017 to 2022. In studies predicting Gleason score, radiomic models differentiated well between Gleason score risk groups and in predicting Gleason grade group upgrading (ROC AUC 0.63-0.89).2,3 Studies predicting adverse pathology also yielded high ROC AUC values between 0.83 and 0.91 for radiomic models, outperforming clinical nomograms in 2 comparative studies.4,5

While these results support the potential use of radiomics in initial risk stratification, final pathology following radical prostatectomy would likely still dictate treatment strategy. As such, the natural progression for clinical integration of radiomics has shifted toward prognostication and prediction of treatment response. Of the 42 articles above, 4 (9.5%) and 6 (14.3%) investigations have highlighted the use of MRI-derived radiomics in predicting post-surgical recurrence and post-radiation failure, respectively. ROC AUCs for these models ranged from 0.71 to 0.73 in the post-surgical and post-radiation cohorts,6,7 values which are comparable to the Memorial Sloan Kettering Cancer Center Pre-Radical Prostatectomy Nomogram and the University of California San Francisco CAPRA (Cancer of the Prostate Risk Assessment) scores for predicting recurrence following surgery. However, given the recency of these studies, only 2 groups have included external validation of their radiomic models, and it is clear that further exploration is required before clinical integration can be considered.

As an MD/PhD scholar at the University of Nebraska Medical Center, my thesis dissertation project centers on increasing applicability of radiomic technologies to prostate cancer prognostication. Working with Dr Michael Baine in the Department of Radiation Oncology has enabled me to shape a project that integrates the technology of imaging-derived radiomics with the clinical care pathway in prostate cancer. Internal development of these radiomic models has yielded promising preliminary findings with high sensitivity in predicting prostate cancer recurrence following radical prostatectomy. Over the coming months, we hope to further enhance the predictive capability of our models by integrating patient demographics and clinical characteristics with the radiomic features. Furthermore, as the project grows, we will continue recruiting other institutions to externally validate our model and its findings.

Overall, prostate cancer radiomics presents as an emerging research field with the potential to offer noninvasive, imaging-based biomarkers for risk stratification and prediction of treatment response. Given the high heterogeneity of prostate cancer, the quantitative characterization of tumor heterogeneity and identification of imaging-based biomarkers may enable disease-tailored treatment planning. Direct application of radiomics to prediction of treatment outcomes, however, remains an ongoing investigation. As these studies mature and reach potential for clinical integration, concerted efforts to standardize methodology and systematically validate these radiomic models must be undertaken.

  1. Shur JD, Doran SJ, Kumar S, et al. Radiomics in oncology: a practical guide. Radiographics. 2021;41(6):1717-1732.
  2. Abdollahi H, Mofid B, Shiri I, et al. Machine learning-based radiomic models to predict intensity-modulated radiation therapy response, Gleason score and stage in prostate cancer. Radiol Med. 2019;124(6):555-567.
  3. Losnegård A, Reisæter LAR, Halvorsen OJ, et al. Magnetic resonance radiomics for prediction of extraprostatic extension in non-favorable intermediate- and high-risk prostate cancer patients. Acta Radiol. 2020;61(11):1570-1579.
  4. Cuocolo R, Stanzione A, Faletti R, et al. MRI index lesion radiomics and machine learning for detection of extraprostatic extension of disease: a multicenter study. Eur Radiol. 2021;31(10):7575-7583.
  5. Ma S, Xie H, Wang H, et al. MRI-based radiomics signature for the preoperative prediction of extracapsular extension of prostate cancer. J Magn Reson Imaging. 2019;50(6):1914-1925.
  6. Li L, Shiradkar R, Leo P, et al. A novel imaging based nomogram for predicting post-surgical biochemical recurrence and adverse pathology of prostate cancer from pre-operative bi-parametric MRI. EBioMedicine. 2021;63:103163.
  7. Zhong QZ, Long LH, Liu A, et al. Radiomics of multiparametric MRI to predict biochemical recurrence of localized prostate cancer after radiation therapy. Front Oncol. 2020;10:731.