Reproducibility of Apparent Diffusion Coefficient and Restriction Spectrum Imaging Restriction Score in the Prostate Across MRI Sessions, Vendors, and Acquisition Settings: a Prospective Study

This prospective study demonstrates that while Apparent Diffusion Coefficient (ADC) shows limited reproducibility across MRI sessions and vendors, the Restriction Spectrum Imaging restriction score maximum value (RSIrs-max) exhibits significantly stronger cross-session reproducibility in prostate cancer detection, even under varying acquisition settings and vendor conditions.

The Gist

The Big Picture: Measuring the "Squeeze" of Prostate Cancer

Imagine the prostate gland is a sponge. When doctors suspect cancer, they use a special type of MRI (a giant magnet camera) to see how water moves inside that sponge. Cancerous tissue is usually tighter and "squeezes" the water more than healthy tissue.

For a long time, doctors have used a standard ruler to measure this "squeeze," called the ADC (Apparent Diffusion Coefficient). However, this study found that this ruler is a bit shaky. If you measure the same spot on the same person on two different days, or if you use two different brands of MRI machines, the number you get can jump around a lot. It's like trying to measure the height of a tree with a ruler that stretches or shrinks depending on the weather or who is holding it.

The researchers wanted to test a new, smarter tool called RSI (Restriction Spectrum Imaging). Specifically, they looked at a number called RSIrs-max, which acts like a "maximum squeeze score." They wanted to see if this new score stayed consistent when the same patient was scanned multiple times, on different machines, or by different hospitals.

The Experiment: The "Double-Check" Test

The researchers gathered 61 men who needed prostate checks. Here is how they set up the test:

1. The "Same-Day" vs. "Different-Day" Test: Each man got two MRI scans. One was part of their normal care, and the second was a research scan done at a different time.
2. The "Brand Switch" Test: Some men were scanned on the same brand of MRI machine (like two different Samsung phones). Others were scanned on completely different brands (like a Samsung vs. an iPhone).
3. The "High-Risk" Group: They paid special attention to men with "unfavorable-histology" cancer (uhPC). Think of this as the "toughest" or most aggressive type of cancer, where getting the measurement right is critical.
4. The "Digital Zoom" Test: They also checked if a specific computer setting (called zero-filled interpolation, which is like a digital zoom or smoothing filter) changed the results.

What They Found: The Old Ruler vs. The New Compass

1. The Old Ruler (ADC) is Wobbly
When they measured the standard "squeeze" (ADC), the results were inconsistent.

• The Analogy: Imagine trying to guess the temperature of a room by looking out the window. If you look from the north side, it feels cool. If you look from the south side, it feels warm. The actual temperature hasn't changed, but your measurement depends on where you stand.
• The Result: When the same man was scanned on different machines, the ADC numbers didn't match up well. It was hard to tell if a change in the number meant the cancer was getting worse or if it was just the machine acting up.

2. The New Compass (RSIrs-max) is Steady
The new "maximum squeeze score" (RSIrs-max) was much more reliable.

• The Analogy: Imagine a GPS that tells you exactly where you are. Even if you switch from a Ford car to a Toyota, or drive on a sunny day vs. a rainy day, the GPS still points to the exact same location.
• The Result: Whether the men were scanned on the same machine or different brands, the RSIrs-max numbers stayed very close to each other. Even in the men with the most aggressive cancer (uhPC), this new score was highly consistent.

3. The "Digital Zoom" Matters
They found that if both scans used the same computer setting (the "digital zoom" or interpolation), the results were even better. If one scan used the setting and the other didn't, the numbers got a bit messy. It's like taking a photo with a filter on one day and without it the next; the colors look different even if the subject is the same.

The Bottom Line

This study is like a quality control check for medical tools.

• The Problem: The current standard tool (ADC) is like a rubber ruler. It changes length depending on the machine or the day, making it hard to track changes over time.
• The Solution: The new tool (RSIrs-max) is like a steel ruler. It gives you the same measurement every time, whether you are using a GE machine or a Siemens machine, and whether you are looking at a mild case or a tough case.

The researchers conclude that while the old method is okay for a quick glance, the new method is much better if you need to track a patient's progress over months or years, or if you need to compare results between different hospitals. They are currently testing this new tool in several large clinical trials to see how it helps with diagnosis and treatment planning.

Technical Summary

Technical Summary: Reproducibility of ADC and RSIrs-max in Prostate MRI

Problem Statement

Diffusion-weighted MRI (DWI) is a cornerstone of prostate cancer detection within multiparametric MRI (mpMRI) protocols. However, the quantitative metric most commonly used, the Apparent Diffusion Coefficient (ADC), suffers from limited reproducibility across different scanner models, vendors, and imaging sites. This variability hinders the reliable use of ADC as a standalone quantitative biomarker for longitudinal monitoring (e.g., active surveillance) and multi-center clinical decision-making. While Restriction Spectrum Imaging (RSI) and its derived Restriction Score (RSIrs) have shown promise in characterizing tissue microstructure and outperforming conventional diffusion metrics in cancer detection, their reproducibility under clinically realistic conditions—specifically across different scan sessions, different scanner vendors, and varying acquisition settings—had not been fully established.

Methodology

This prospective study (NCT04992728) enrolled 61 male participants (median age 69) with suspected or known prostate cancer between August 2022 and January 2026. Each participant underwent two separate MRI examinations:

1. Exam 1: A standard-of-care mpMRI protocol (PI-RADS v2.1 compliant) including RSI, performed either at the University of California San Diego (UCSD) or a community imaging center.
2. Exam 2: A research-dedicated session performed at UCSD on a different scanner, including RSI and high-resolution T2-weighted imaging.

Data Processing and Analysis:

• Lesion Selection: MRI-visible lesions were contoured on T2-weighted images. For patients with multiple lesions, the "index lesion" was selected based on Grade Group (GG), cribriform status, or PI-RADS score.
• Metrics: Mean ADC and the maximum RSI restriction score (RSIrs-max) were measured within these lesion contours.
• Subgroups: Analyses were stratified by:
  • Vendor Consistency: Same-vendor vs. cross-vendor scanner pairs.
  • Histology: Participants with unfavorable-histology prostate cancer (uhPC), defined as GG≥3 or intermediate-grade disease with cribriform morphology.
  • Interpolation Settings: A subset of 20 participants underwent scans reconstructed with and without zero-filled interpolation to assess the impact of this specific acquisition parameter.
• Statistical Approach: Reproducibility was quantified using Pearson correlation coefficients with 10,000 bootstrap resamples to estimate 95% confidence intervals (CI). The study utilized scatter plots to visualize agreement between paired measurements.

Key Results

The study evaluated 58 participants with controllable MRI-visible lesions (26 of whom had uhPC).

1. Overall Reproducibility:

• ADC: Showed limited reproducibility across all visible lesions with a correlation of 0.55 (95% CI: 0.23–0.76).
• RSIrs-max: Demonstrated significantly stronger reproducibility with a correlation of 0.83 (95% CI: 0.72–0.90).

2. Vendor Comparisons:

• Same-Vendor: ADC correlation was 0.76; RSIrs-max correlation was 0.88.
• Cross-Vendor: ADC correlation dropped significantly to 0.31 (95% CI: -0.07–0.62), indicating poor consistency between manufacturers. In contrast, RSIrs-max maintained strong reproducibility at 0.79 (95% CI: 0.65–0.89).

3. Unfavorable-Histology Prostate Cancer (uhPC):

• In this high-risk subgroup, ADC reproducibility was poor (correlation 0.42).
• RSIrs-max showed exceptional stability, with correlations of 0.90 (same vendor) and 0.92 (cross-vendor).

4. Impact of Zero-Filled Interpolation:

• Inconsistent application of zero-filled interpolation (one scan with, one without) resulted in an RSIrs-max correlation of 0.73.
• When interpolation was applied consistently to both scans, the correlation improved to 0.89.

Key Contributions

• Quantitative Benchmarking: This study provides the first prospective, cross-session, and cross-vendor comparison of ADC versus RSIrs-max reproducibility in prostate MRI.
• Validation of RSIrs-max: It establishes that RSIrs-max is a more robust quantitative biomarker than mean ADC, maintaining high reproducibility even when scanning conditions (vendor, session) change.
• High-Risk Subgroup Analysis: It specifically validates the utility of RSIrs-max in patients with unfavorable-histology prostate cancer, a group where reliable longitudinal monitoring is critical but historically difficult due to ADC variability.
• Technical Optimization: The study identifies zero-filled interpolation as a critical preprocessing variable, demonstrating that consistent application of this setting significantly enhances the reproducibility of RSI metrics.

Significance and Claims

The authors conclude that while ADC remains a standard component of qualitative PI-RADS interpretation, its variability across sessions and vendors limits its utility as a precise quantitative biomarker for longitudinal tracking. In contrast, RSIrs-max demonstrates strong between-session reproducibility that is comparable to intra-session repeatability (scans minutes apart) and remains robust across different manufacturers.

The paper posits that the superior stability of RSIrs-max makes it a viable candidate for:

• Longitudinal monitoring in active surveillance and treatment response assessment.
• Multi-center clinical trials where scanner heterogeneity is a confounding factor.
• Risk stratification in patients with unfavorable-histology disease.

The authors note that these findings support the ongoing evaluation of RSIrs-max in prospective trials for initial diagnosis, radiotherapy targeting, and treatment response assessment. They emphasize that while RSIrs-max is inherently more sensitive to noise (being a maximum value rather than a mean), it still outperforms mean ADC in reproducibility, suggesting the underlying RSI model effectively mitigates sources of variability that affect standard ADC calculations.