A recent multi-reader clinical study has demonstrated that the Prostate Simplified MRI Anatomy Recognition and Training (P-SMART) method significantly improves interobserver agreement for prostate lesion localisation compared to the conventional Prostate Imaging-Reporting and Data System (PI-RADS) approach. The evaluation, which analysed inter-reader consistency across 156 regions of interest (ROIs) on T2-weighted magnetic resonance images, found that the P-SMART approach achieved a Fleiss’ kappa coefficient of 0.571, compared to 0.399 for the current standard PI-RADS method. This improvement was statistically significant (p < 0.001) and consistent across various reader experience levels and prostate gland sizes, offering a potential path forward for standardising the reporting of prostate MRI findings in clinical practice.
Prostate cancer management relies heavily on MRI for diagnosis, staging, and monitoring. Accurate documentation of lesion location within the prostate gland—typically divided into the apex, mid-gland, and base—is essential for biopsy planning, active surveillance monitoring, and surgical decision-making. Despite its importance, the current PI-RADS guidance for defining these levels lacks granular, systematic instructions. The sector map often utilised in practice assumes a general proportionality of the gland’s long axis, which may fail in patients with benign prostatic hyperplasia (BPH) or other anatomical variations. This ambiguity has led to documented variability in how clinicians interpret and communicate the anatomical location of prostate lesions.
To investigate whether a more structured, anatomy-based approach could improve consistency, the P-SMART study researchers conducted a two-phase agreement evaluation. Using a publicly available dataset of 40 prostate MRI cases, the study enrolled 43 radiologist participants. These readers were asked to document the location of ROIs—specifically identifying them as either the apex, mid-gland, or base—first using the existing PI-RADS methodology and then, following a three-week washout period and an educational intervention, using the novel P-SMART approach. The study design aimed to isolate the method of level assignment as the primary variable by using identical MRI cases in both phases while ensuring that the order of presentation was rearranged to mitigate recall bias.
The P-SMART method relies on clear anatomical landmarks to delineate the prostate regions. The approach involves identifying four specific landmarks on sagittal T2-weighted MRI images: the bladder-prostate angle anteriorly, the seminal vesicle-prostate angle posteriorly, and the most inferior portions of the prostatic apex both anterior and posterior to the urethra. Once these landmarks are identified, the method employs a three-step process—identifying the points, dividing the anterior and posterior lines based on defined proportions, and joining these points to form boundaries. This systematic framework is designed to adapt to normal variations in prostate morphology, such as those caused by median lobe hypertrophy, ensuring that the peripheral zone is consistently represented at the base.
Statistical analysis of the interobserver agreement was performed using Fleiss’ kappa coefficient, a measure that accounts for agreement beyond what might be expected by chance. The researchers utilised bootstrap resampling with 1,000 iterations to compare the kappa distributions of the two methods, applying paired t-tests to compute p-values and 95% confidence intervals. The study also conducted subgroup analyses to evaluate the performance of the methods stratified by reader expertise and prostate size. Readers were classified as experts if they had independently read more than 1,000 prostate MRI studies, and glands were classified as large or small based on a median volume of 60 mL within the dataset.
The study’s findings indicated that the P-SMART approach outperformed the PI-RADS method in all assessed cohorts. For expert readers, the kappa coefficient improved from 0.435 to 0.663, while for non-expert readers, it increased from 0.393 to 0.530. Similar performance gains were observed when stratifying by gland size; for large prostates, agreement increased from 0.325 to 0.518, and for small prostates, it rose from 0.461 to 0.620. These results suggest that by reducing reliance on subjective interpretation, a standardised protocol based on anatomical landmarks can enhance the reproducibility of prostate MRI reporting, which is a critical factor for the reliability of clinical communication.
A significant portion of the discussion addressed the complexities of prostate morphology. The researchers noted that large prostates, particularly those affected by BPH, often present the greatest challenge for level localisation because the transition zone can extend cranially relative to the peripheral zone. While P-SMART demonstrated improvement in these challenging cases, the study acknowledged that agreement remained lower for larger glands than for smaller ones, potentially because the required anatomical landmarks are more difficult to define when the gland structure is significantly altered by hypertrophy. This observation suggests that even structured approaches have inherent limitations tied to patient anatomy.
The authors compared their results to previous studies, such as those by Greer et al., which highlighted the dramatic decline in concordance when the PI-RADS sector map was introduced for lesion localisation. The high degree of variability in interpreting conventional maps often leads to miscommunication, which can have downstream consequences for biopsy targeting and surgical planning, such as potentially compromising nerve-sparing efforts during prostatectomy. By establishing a more reproducible framework, P-SMART attempts to address these issues, although the researchers maintained that their method is intended to complement current reporting conventions rather than replace them entirely.
Several limitations were discussed, including the use of artificial ROIs rather than real clinical lesions. While this approach effectively isolated level assignment as the variable of interest, it remains unclear how the P-SMART method would perform with actual lesions of varying size and conspicuity, which may span multiple levels. Furthermore, the researchers acknowledged the potential for instructional bias, as the P-SMART training included an eight-minute instructional video, whereas the PI-RADS phase relied on standard definitions without a comparable structured video. This difference was cited as inherent to comparing a standardised protocol with routine clinical practice, where no such standardised video protocol for PI-RADS exists.
Additional limitations included the reliance on a historic dataset from 2011 to 2012, which may not fully represent the image quality achievable with modern MRI hardware and contemporary acquisition protocols. The study also excluded two participants due to fundamental orientation errors identified during a post-hoc quality assurance review, an action the authors noted could introduce selection bias. However, the study provided strong evidence that a structured, anatomy-based approach significantly improves reproducibility compared to existing practices, regardless of the reader’s prior experience level or the size of the gland being evaluated.
Looking toward future implementation, the researchers suggest that the P-SMART method could potentially be integrated into automated reporting workflows and artificial intelligence tools. As AI-based solutions for gland volume calculation and lesion identification become increasingly prevalent in clinical radiology, incorporating structured anatomical landmarks could improve the consistency of automatically generated reports. However, the authors emphasised that prospective clinical validation is required to determine whether improved interobserver agreement translates into measurable clinical benefits, such as more accurate biopsy targeting or improved surgical outcomes.
Ultimately, while the P-SMART approach does not eliminate all subjectivity, it provides a measurable advancement in the standardisation of prostate MRI anatomy recognition. The moderate agreement reached even with the novel method suggests that prostate level classification will always involve a degree of interpretive judgment based on individual anatomy. Nevertheless, the consistent performance across both experts and less experienced readers underscores the potential of the P-SMART method to serve as a valuable tool for radiologists aiming to improve the reliability and clarity of their communications with urologists and other members of the multidisciplinary team.
Subject of Research: Evaluation of the P-SMART anatomical landmark-based approach for improving interobserver agreement in prostate MRI lesion localisation.
Article Title: Prostate: simplified MRI anatomy recognition and training—the P-SMART study: a novel approach to defining prostatic apex, mid-gland and base
Article References: Withey, S. J., Gutierrez, P., Ponsiglione, A. M., Stanzione, A., Zawaideh, J. P., Brembilla, G., Avesani, G., Hamm, C. A., Dias, A. B., Woo, S., Padhani, A. R., Løgager, V., Ponsiglione, A., P-SMART Collaborators, Abreu-Gomez, J., Agrotis, G., Alessandrino, F., Alkhalifah, F., Arita, Y., … Yang, Y. H. (2026). Prostate: simplified MRI anatomy recognition and training—the P-SMART study: a novel approach to defining prostatic apex, mid-gland and base. Insights into Imaging, 17(1), Article 281. https://doi.org/10.1186/s13244-026-02459-2
Image Credits: AI Generated
DOI: 10.1186/s13244-026-02459-2
Keywords: Prostate MRI, PI-RADS, P-SMART, Radiology, Interobserver agreement, Fleiss' kappa, Prostate cancer, Biopsy planning, Active surveillance, Medical imaging, Anatomical landmarks, Prostate anatomy, Diagnostic standardisation, Clinical radiology, MRI interpretation, Urogenital imaging, Lesion localisation
Cite Scienmag News
Ophelia Keating. (November 18, 2026). Anatomy-Based Prostate MRI Standardisation Improves Reader Interpretation Agreement. Scienmag. https://scienmag.com/anatomy-based-prostate-mri-standardisation-improves-reader-interpretation-agreement/
Ophelia Keating. "Anatomy-Based Prostate MRI Standardisation Improves Reader Interpretation Agreement." Scienmag, 18 November 2026, https://scienmag.com/anatomy-based-prostate-mri-standardisation-improves-reader-interpretation-agreement/. Accessed 18 November 2026.
Ophelia Keating. "Anatomy-Based Prostate MRI Standardisation Improves Reader Interpretation Agreement." Scienmag. November 18, 2026. https://scienmag.com/anatomy-based-prostate-mri-standardisation-improves-reader-interpretation-agreement/








