The Importance of Quality in Prostate MRI

Deep Learning-Based Interpretable AI for Prostate T2W MRI Quality Evaluation

RATIONALE AND OBJECTIVES

Prostate MRI quality is essential in guiding prostate biopsies. However, assessment of MRI quality is subjective with variation. Quality degradation sources exert varying impacts based on the sequence under consideration, such as T2W versus DWI. As a result, employing sequence-specific techniques for quality assessment could yield more advantageous outcomes. This study aims to develop an AI tool that offers a more consistent evaluation of T2W prostate MRI quality, efficiently identifying suboptimal scans while minimizing user bias.

MATERIALS AND METHODS

This retrospective study included 1046 patients from three cohorts (ProstateX [n = 347], All-comer in-house [n = 602], enriched bad-quality MRI in-house [n = 97]) scanned between January 2011 and May 2022. An expert reader assigned T2W MRIs a quality score. A train-validation-test split of 70:15:15 was applied, ensuring equal distribution of MRI scanners and protocols across all partitions. T2W quality AI classification model was based on 3D DenseNet121 architecture using MONAI framework. In addition to multiclassification, binary classification was utilized (Classes 0/1 vs. 2). A score of 0 was given to scans considered non-diagnostic or unusable, a score of 1 was given to those with acceptable diagnostic quality with some usability but with some quality distortions present, and a score of 2 was given to those considered optimal diagnostic quality and usability. Partial occlusion sensitivity maps were generated for anatomical correlation. Three body radiologists assessed reproducibility within a subgroup of 60 test cases using weighted Cohen Kappa.

RESULTS

The best validation multiclass accuracy of 77.1% (121/157) was achieved during training. In the test dataset, multiclassification accuracy was 73.9% (116/157), whereas binary accuracy was 84.7% (133/157). Sub-class sensitivity for binary quality distortion classification for class 0 was 100% (18/18), and sub-class specificity for T2W classification of absence/minimal quality distortions for class 2 was 90.5% (95/105). All three readers showed moderate to substantial agreement with ground truth (R1-R3 κ = 0.588, κ = 0.649, κ = 0.487, respectively), moderate to substantial agreement with each other (R1-R2 κ = 0.599, R1-R3 κ = 0.612, R2-R3 κ = 0.685), fair to moderate agreement with AI (R1-R3 κ = 0.445, κ = 0.410, κ = 0.292, respectively). AI showed substantial agreement with ground truth (κ = 0.704). 3D quality heatmap evaluation revealed that the most critical non-diagnostic quality imaging features from an AI perspective related to obscuration of the rectoprostatic space (94.4%, 17/18).

CONCLUSION

The 3D AI model can assess T2W prostate MRI quality with moderate accuracy and translate whole sequence-level classification labels into 3D voxel-level quality heatmaps for interpretation. Image quality has a significant downstream impact on ruling out clinically significant cancers. AI may be able to help with reproducible identification of MRI sequences requiring re-acquisition with explainability.

Is imaging accurate enough to detect index lesion in prostate cancer? Analysis of the performance of MRI and other imaging modalities

Prostate imaging techniques have progressed across the years allowing for a better detection and characterization of prostate cancer (PCa) lesions. These advancements have led to the possibility to also improve and tailor the treatments on the most aggressive lesion, defined as Index Lesion (IL), to reduce morbidity. The IL is, indeed, considered as the entity which encompass the most aggressive features in prostate cancer disease. Multiparametric magnetic resonance imaging (mpMRI) has emerged as the suggested tool to detect the disease and plan treatments, including those under investigation such as focal therapy (FT). Our review aimed to query the literature on the ability of mpMRI in IL detection and to explore the future perspectives in PCa IL diagnosis. A review of the literature was performed from January 2010 to July 2023. All studies investigating the performance of mpMRI and other main imaging techniques able to detect the IL were assessed and evaluated. mpMRI performs well in the detection of IL with a sensitivity which reaches 71% to 94% among the different studies. However, mpMRI seems to have limited sensitivity in the detection of small tumours (<0.5 mL) and low-grade histology lesions. To overcome these limitations other diagnostic imaging techniques have been proposed. Multiparametric Ultrasound has shown results comparable to mpMRI while detecting 4.3% fewer clinically significant PCa (P=0.042). Positron emission tomography-based modalities using PSMA seems to have higher sensitivity than mpMRI, being able to yield from 13.5% to 18.2% additional cancers. MRI has emerged as the recommended tool since most of the IL can be easily identified, and is the imaging of choice while selecting patients for FT. Other imaging modalities has been proposed to improve PCa lesions detection, but results need to be confirmed by ongoing randomized controlled trial.

Quality of T2-weighted MRI re-acquisition versus deep learning GAN image reconstruction: A multi-reader study

PURPOSE

To evaluate CycleGAN's ability to enhance T2-weighted image (T2WI) quality.

METHOD

A CycleGAN algorithm was used to enhance T2WI quality. 96 patients (192 scans) were identified from patients who underwent multiple axial T2WI due to poor quality on the first attempt (RAD1) and improved quality on re-acquisition (RAD2). CycleGAN algorithm gave DL classifier scores (0-1) for quality quantification and produced enhanced versions of QI1 and QI2 from RAD1 and RAD2, respectively. A subset (n = 20 patients) was selected for a blinded, multi-reader study, where four radiologists rated T2WI on a scale of 1-4 for quality. The multi-reader study presented readers with 60 image pairs (RAD1 vs RAD2, RAD1 vs QI1, and RAD2 vs QI2), allowing for selecting sequence preferences and quantifying the quality changes.

RESULTS

The DL classifier correctly discerned 71.9 % of quality classes, with 90.6 % (96/106) as poor quality and 48.8 % (42/86) as diagnostic in original sequences (RAD1, RAD2). CycleGAN images (QI1, QI2) demonstrated quantitative improvements, with consistently higher DL classifier scores than original scans (p < 0.001). In the multi-reader analysis, CycleGAN demonstrated no qualitative improvements, with diminished overall quality and motion in QI2 in most patients compared to RAD2, with noise levels remaining similar (8/20). No readers preferred QI2 to RAD2 for diagnosis.

CONCLUSION

Despite quantitative enhancements with CycleGAN, there was no qualitative boost in T2WI diagnostic quality, noise, or motion. Expert radiologists didn't favor CycleGAN images over standard scans, highlighting the divide between quantitative and qualitative metrics.

Optimization of non-endorectal prostate MR image quality using PI-QUAL: A multidisciplinary team approach

PURPOSE

To evaluate the utility of the PI-QUAL score in assessing protocol changes aimed to improve image quality from a non-endorectal coil prostate MR imaging protocol during a 9-month quality improvement (QI) project and to quantify the inter-reader agreement of PI-QUAL scores between radiologists, technologists, and physicists.

METHODS

This retrospective study audited 1,012 multiparametric prostate MRI examinations as part of a national QI project according to the PI-QUAL standard. PI-QUAL scores were used to inform MR protocol changes. Following the project, 4 radiologists, 2 technologists, and 1 medical physicist collectively audited an additional set of 150 examinations to identify statistical improvements in image quality using the two-tailed Wilcoxon rank sum test. The improvements due to individual protocol changes were assessed among subsets of the 1,012 examinations which compared examinations occurring before and after the isolated protocol change. Inter-reader variability was assessed using the percent majority agreement and the average standard deviation of PI-QUAL scores between evaluators.

RESULTS

During this QI project, PI-QUAL scores improved from 3.67 ± 0.75 to 4.16 ± 0.59 (p < 0.01) after implementing a series of protocol changes. Among a subset of 451 cases, we found that adopting R/L rather than A/P phase encoding reduced distortion in diffusion-weighted imaging (DW) from 21.6% (41/190 A/P phase encoded cases) to 11.5% (30/261 R/L phase encoded cases) (p < 0.01). Similarly, in the same 451 cases, adopting R/L phase encoding in T2WI reduced breathing motion artifacts from 34.6% (94/272 A/P phase encoding cases) to 12.8% (23/179 R/L phase encoding cases) (p < 0.01). DWI wraparound artifact was mitigated by employing a full-pelvis shim and enabling the abdomen shim option. The occurrence of low signal-to-noise ratio was reduced from 19.4% (19/98 cases without a weight-based threshold) to 6.3% (10/160) by instituting a weight-based threshold for using an endorectal coil (p < 0.01). The percent majority agreement was similar between radiologists, technologists and physicists, and all evaluators combined (72%, 77%, and 67%, respectively).

CONCLUSIONS

PI-QUAL can evaluate image quality changes resulting from protocol optimizations at both the exam- and series-levels. With training, radiologists, technologists, and physicists can perform PI-QUAL scoring with similar performance. Broadening the scope of the quality improvement team can result in meaningful and lasting change.

Reconciling discordance between PI-RADS 4 lesions and targeted biopsy: Early experience of a multidisciplinary quality improvement protocol with PI-RADS 4 subcategorization

PURPOSE

PI-RADS 4 lesions are considered to have a "high" likelihood of clinically-significant prostate cancer (csPCa). However, patients undergoing targeted biopsy have a range of histologic findings. Understanding discordant cases is critical to improve diagnostic accuracy and inform subsequent management. We studied early findings from implementation of a multidisciplinary Quality Improvement (QI) protocol for reconciling discordance and evaluate the potential heterogeneity of PI-RADS 4.

METHODS

Patients with mpMRI PI-RADS 4 lesions undergoing fusion-targeted biopsy from January 2017 to May 2021 were retrospectively reviewed. The discordant targeted biopsy pathology (benign/GG1) was evaluated utilizing a QI protocol and all lesions were subcategorized based on ADC values. Positive Predictive Value (PPV) for PI-RADS 4 lesions overall and the Cancer Detection Rate (CDR) for subcategorized lesions were calculated.

RESULTS

248 patients with 286 lesions were reviewed. Prior to re-review, PI-RADS 4 PPV for ≥ GG1 and ≥ GG2 lesions were 0.55 and 0.34, increasing to 0.67 and 0.43 following reconciliation. Lesion subcategorization based on ADC value as higher suspicion (4+) and lower suspicion (4-) resulted in 158 and 117 lesions, with reverse-fusion analysis revealing that 61% and 17% of lesions contained csPCa, respectively. Subgroup analysis among PI-RADS 4+ lesions led to an increase in the CDR to 75% and 61% for ≥ GG1 and ≥ GG2.

CONCLUSION

Use of multidisciplinary QI protocol to review discordance cases of PI-RADS 4 improves diagnostic accuracy and guides subsequent management. Our findings highlight the known heterogeneity of this category with reference to csPCa CDR, suggesting the potential value of PI-RADS 4 subcategorization.