Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy: pre-declared margins, patient-level acceptance thresholds and the incremental predictive value of DVH indices

arXiv:2609.14028 · physics.med-ph · Submitted 2026-09-12 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy".

Mira: Here's the gist of the paper. Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy: pre-declared margins, patient-level acceptance thresholds and the incremental predictive value of DVH indices is about comparing dose-volume histogram (DVH) indices derived from deep learning prostate contours against expert manual delineation to assess dosimetric equivalence and their predictive value for patient-reported outcomes like the International Prostate Symptom Score (IPSS).

Kai: This study investigates whether automatic contours produce DVH errors comparable to human interobserver variability and whether these indices can predict changes in IPSS over time Dosimetric Equivalence Testing The core question addressed is whether automatic contours produce DVH errors comparable to human interobserver variability, and if these indices can predict patient-reported toxicity.

Mira: Equivalence is defined as an automatic contour being acceptable when the dosimetric error it produces is no larger than the error two experts already produce between themselves The study compared indices from expert manual delineation with deterministic and Bayesian nnUNet models against margins set from CT contouring IOV.

Kai: Cohort-Level Equivalence Results All cohort-level comparisons of DVH indices were declared equivalent, with no interval consuming more than 44 % of its equivalence margin.

Title and authors: Mira: The paper sets up this comparison by defining equivalence in a specific way. They say an automatic contour is acceptable if the dose error it produces isn't bigger than the error two experts already make when they draw it themselves, which is their way of setting that initial margin for acceptable deviation > <ref:2609.14028#pg3>.

Kai: So they aren't just saying "it looks okay"; they’re using the existing human disagreement as the benchmark for what an automatic contour should be allowed to deviate from. They compared these indices from expert manual delineation against models like deterministic and Bayesian nnUNet, checking them against margins set from how well people were contouring during CT scanning > <ref:2609.14028#pg3>.

Lev: That makes sense for the methodology. It sets up a controlled test where the error tolerance is based on human interobserver variability, which is a pretty realistic way to define what "good enough" means in this context.

Kai: And then they move on to look at individual patients. They figured if it’s not about the whole group, we need a patient-specific rule for when an automatic contour is trustworthy > <ref:2609.14028#pg1>.

Mira: Which brings us to the concept of patient-level quality thresholds. They used a logistic regression to figure out what segmentation quality score would make an index reliable at a ninety percent probability, and they found that this threshold actually depends on the actual size of the prostate > <ref:2609.14028#pg1>.

Lev: That’s important—it means you can't just use one fixed rule for everyone; the system has to adapt based on how much tissue there is to account for that variability in error > <ref:2609.14028#pg1>.

Kai: So, if we look at how the paper summarizes this whole setup, it’s about testing if the dose metrics from these automatic contours align with human expertise and establishing a rule for when those metrics are trustworthy on a per-patient basis > <ref:2609.14028#pg2>.

Mira: It’s essentially asking two big questions: first, is the automated output dosimetrically equivalent to expert manual drawings, and second, what is the threshold for accepting that equivalence for an individual patient > <ref:2609.14028#pg3>.

The paper's summary: Kai: Moving into what the study actually found, the summary is pretty straightforward. They compared expert manual contours with deterministic and Bayesian nnUNet models against those pre-declared equivalence margins from contouring reproducibility > <ref:2609.14028#pg3>.

Mira: The main finding on equivalence across all cohort comparisons was that they were all declared equivalent, meaning no interval consumed more than forty-four percent of its established margin for being considered acceptable > <ref:2609.14028#pg3>.

Lev: That suggests a good level of agreement between the different segmentation sources when measured against those human-defined error tolerances. It shows that the models are capturing the core dosimetric characteristics well enough to pass that initial test on average > <ref:2609.14028#pg3>.

Kai: But they also found some differences in individual agreement, depending on which specific dose-volume histogram index you were looking at and which automatic segmentation model you used, with equivalence rates ranging from fifty-five point nine percent to eighty-nine point seven percent > <ref:2609.14028#pg3>.

Mira: And they noted that the deterministic network tended to overestimate every DVH index by three point three four percentage points of prescription on the D90, and about zero point nine seven to one point four one points of volume on those three Vx indices > <ref:2609.14028#pg3>.

Lev: So, even though the overall comparison passed the equivalence test, there’s a clear bias in how that deterministic network estimates certain metrics compared to the expert manual ones > <ref:2609.14028#pg3>.

Kai: That deterministic network overestimation is something we need to keep in mind when we look at real-world application, because it shows where the model has specific systematic errors, not just random noise > <ref:2609.14028#pg3>.

The paper's improvements: Mira: Now they move into suggesting improvements for the system itself, which is a key part of their discussion. They’re focused on making the system smarter about when it’s actually reliable for a specific patient > <ref:2609.14028#pg1>.

Kai: The improvements suggested are quite detailed. First, they suggest setting an adaptive equivalence margin that adjusts based on the actual prostate volume of the patient > <ref:2609.14028#pg1>.

Lev: That means instead of one fixed error limit for everyone, you dynamically change the acceptable error based on how big or small the target area is, which makes a lot more sense clinically > <ref:2609.14028#pg1>.

Mira: They also propose using volume-dependent quality scoring. They suggest fitting a logistic regression model to determine that quality score as a predictor of equivalence at a ninety percent probability, and this score is derived by comparing eight candidate metrics spanning overlap, distance and volumetric families as predictors of E equals one Delta less than or equal to delta > Improvement two.

Kai: So they’re not just looking at one metric; they’re building a composite quality score using eight different measures of overlap and distance, which sounds like it captures more nuance about the segmentation quality > Improvement two.

Lev: That level of input for the quality score is necessary if you want that predictive power to hold up, because you need those metrics to cover all aspects of how well the contour aligns with reality > Improvement two.

Mira: They also suggest a directional error analysis. Instead of just saying the error is large, they want to calculate the signed index difference regressed on the signed relative volume error so we know precisely in which direction and by how much it departs from its manual counterpart when it does deviate > Improvement three <ref:2609.14028#pg3>.

Kai: That’s useful because if you know *how* it’s wrong—is it overestimating or underestimating volume—that information is much more actionable for a clinician than just an unsigned error metric like Dice > Improvement three <ref:2609.14028#pg3>.

Conclusion: Mira: To wrap up, the conclusion is pretty definitive about the predictive value. They found that across all learners and regardless of whether they were looking at the segmentation source or looking at a time horizon, these dose-volume histogram indices did not provide any significant predictive power for changes in urinary morbidity, like IPSS > <ref:2609.14028#pg1>.

Kai: It seems the main takeaway is that while the deep learning contours reproduce the expert dose-volume indices quite closely—so source doesn't change what they predict—in this cohort, that prediction is essentially just as good as the baseline clinical performance > <ref:2609.14028#pg1>.

Lev: And when we look at the room available for error, even under the most conservative floor, there’s still at least one point one seven points of removable error at every time horizon, and the largest shift any block produces is just zero point zero five five points > <ref:2609.14028#pg1>.

Mira: So they conclude that the reliability question and the predictive question are actually separate issues. You can answer both by setting the right margins and criteria beforehand, rather than trying to read a single result off the data > <ref:2609.14028#pg3>.

Kai: It’s about using those predefined rules, not just chasing some prediction that doesn't actually add anything clinically meaningful right now. That’s the big picture here for this paper on "Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy: pre-declared margins, patient-level acceptance thresholds and the incremental predictive value of DVH indices".

Lev: It confirms that while the segmentation is close enough to human experts, those specific dose metrics don't offer a measurable advantage over what we already expect from baseline performance > <ref:2609.14028#pg1>.

Université Laval Department of Physics Engineering and Optics Radiation Oncology Service Department of Specialized Medicine Centre Intégré de Cancérologie Hôpital de l’Enfant-Jésus Centre Hospitalier Universitaire Québec Université Laval Research Center of the CHU Québec Université Laval Faculty of Medicine

physics.med-ph

Submitted: 2026-09-12

Updated: 2026-11-10

Comments: Revised based on comments following submission to Radiotherapy and Oncology

Code: https://github.com/lb0918/Dosimetric_equivalence_prostate_

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 83/100

The gist: Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy: pre-declared margins…

Key concepts

Dosimetric Equivalence Testing
This tests if an automatic contour's resulting dose metrics are as good as those from human experts. Equivalence is established when the error produced by the automatic contour is less than the error between two independent human experts.
Patient-Level Quality Thresholds
This determines at what level of segmentation quality an automatic contour becomes reliable for a single patient. A logistic regression model was used to find the quality score threshold where an index derived from the automatic contour is considered trustworthy.
Incremental Predictive Value of DVH Indices
This assesses whether using DVH indices derived from deep learning contours can predict changes in patient outcomes, such as the IPSS. The study found no significant predictive power for IPSS, meaning these indices do not offer extra clinical insight beyond what is already known.

Terminology

Summary

Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy: pre-declared margins, patient-level acceptance thresholds and the incremental predictive value of DVH indices is about comparing dose-volume histogram (DVH) indices derived from deep learning prostate contours against expert manual delineation to assess dosimetric equivalence and their predictive value for patient-reported outcomes like the International Prostate Symptom Score (IPSS). p. 1

Where the paper says it

Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy: pre-declared margins, patient-level acceptance thresholds and the incremental predictive value of DVH indices Louis-Bernard St-Cyra,d , Anne Saint-Laurente , José Angel Lesteiro-Tejedaa,d , Sylviane Aubinc , Marie-Claude Lavalléec , Luc Beaulieua,c,d , Éric Vigneaultb,d , André-Guy Martinb,d , François-Olivier Fabib,c,1 and Louis Archambaulta,c,d,∗,1 aUniversité Laval, Département de physique, génie physique et d’optique, Québec, QC, Canada b Radiation Oncology Service, Department of Specialized Medicine, Centre Intégré de Cancérologie (CIC), Hôpital de l’Enfant-Jésus, Centre Hospitalier Universitaire (CHU) de Québec-Université Laval, Québec, QC, Canada cService of Radio-oncology, CHU de Québec-Université Laval, Québec, QC, Canada d Oncology division, Research Center of the CHU de Québec-Université Laval, Québec, QC, Canada eUniversité Laval, Faculté de médecine, Québec, QC, Canada A R T I C L E I N F O Keywords: Prostate brachytherapy Auto-segmentation Equivalence testing Dose-volume histogram Patient-reported outcomes Deep learning Toxicity prediction A B S T R A C T Background and purpose: Dose Volume Histogram (DVH) indices remain the main doseeffect metrics for toxicity prediction, but they depend on how contours were made. Interobserver variability (IOV) is unavoidable and clinically accepted, so the question for automatic segmentation addresses equivalence: do automatic contours produce DVH errors comparable to human IOV, and can these indices predict patient-reported toxicity?

This study investigates whether automatic contours produce DVH errors comparable to human interobserver variability and whether these indices can predict changes in IPSS over time

Dosimetric Equivalence Testing

The core question addressed is whether automatic contours produce DVH errors comparable to human interobserver variability, and if these indices can predict patient-reported toxicity. Equivalence is defined as an automatic contour being acceptable when the dosimetric error it produces is no larger than the error two experts already produce between themselves The study compared indices from expert manual delineation with deterministic and Bayesian nnUNet models against margins set from CT contouring IOV. p. 2

Where the paper says it

Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy When that link is made, the questions asked are whether the dose-volume histogram (DVH) indices derived from automatic contours differ significantly from the expert ones, and whether they remain acceptable for treatment planning[24, 3, 40, 21]. Equivalence is rarely the question, although it is the one that matters: an automatic contour is acceptable when the dosimetric error it produces is no larger than the error two experts already produce between themselves[11, 26].

Cohort-Level Equivalence Results

All cohort-level comparisons of DVH indices were declared equivalent, with no interval consuming more than 44 % of its equivalence margin. Individual agreement was weaker with equivalence rates from 55. p. 1

Where the paper says it

Results: All cohort-level comparisons of DVH indices were declared equivalent, with no interval consuming more than 44 % of its equivalence margin. Individual agreement was weaker with equivalence rates from 55.

9 % to 89.7 %, depending on the DVH index and automatic segmentation model. p. 1
Where the paper says it

Individual agreement was weaker with equivalence rates from 55. 7 %, depending on the DVH index and automatic segmentation model.

The deterministic network over-estimated every DVH index, by 3. p. 4
Where the paper says it

Cohort-level equivalence All eight comparisons were declared equivalent, with 𝑝TOST < 10−19 throughout and the two automatic contour sources are offset in opposite directions (Table 2A, Fig. The deterministic network over-estimated every DVH index, by 3.

34 percentage points of prescription on D90 and by 0. p. 4
Where the paper says it

Dosimetric equivalence of deep learning prostate contours after LDR brachytherapy DVH indices margins were fixed before any difference was examined and set at one standard deviation of the inter-observer contouring noise floor. Only the contouring arm of published uncertainty and only CT-derived anchors were admissible, this cohort being entirely post-implant CT[30, 17, 11, 47, 39]: 10 percentage points of prescription for 𝐷90, 3 percentage points of volume for 𝑉100 and 4 for 𝑉150 and 𝑉200, the last being extrapolated since no published figure is available (Supplementary S3).

97 to 1.41 points of volume on the three Vx indices. p. 4
Where the paper says it

34 percentage points of prescription on 𝐷90 and by 0. 41 points of volume on the three 𝑉𝑥 indices, L.

Patient-Level Quality Thresholds

For a single patient, the question is one of classification: above what segmentation quality (threshold) is a DVH index derived from an automatic contour reliable. A logistic regression of E on a segmentation quality score was fitted for each DVH index at their respective margins and read at an equivalence probability p∗ = 0. p. 4

Where the paper says it

Patient-level quality thresholds For a single patient the question is one of classification: above what segmentation quality (threshold) is a DVH index derived from an automatic contour reliable? A logistic regression of E on a segmentation quality score was fitted for each DVH index at their respective margins and read at an equivalence probability 𝑝 ∗ = 0.

90. The threshold model chosen between an additive form, Dice with volume, and one carrying their interaction. p. 4
Where the paper says it

The segmentation quality score is the Dice coefficient, which is established rather than assumed, by comparing eight candidate metrics spanning overlap, distance and volumetric families as predictors of E (Supplementary S3). The clinically recorded prostate volume was then added, and the threshold model chosen between an additive form, Dice with volume, and one carrying their interaction, which turns the threshold into a curve in volume.

Incremental Predictive Value of DVH Indices

The DVH indices panel provides no significant predictive power for IPSS, irrespective of segmentation source and horizon. The largest improvement declared by bootstrap intervals was 0. p. 1

Where the paper says it

No DVH block improved IPSS prediction at any horizon. The largest improvement declared by bootstrap intervals was 0.

12 IPSS points, well below the minimal clinically important difference, across all learners. p. 1
Where the paper says it

The largest improvement declared by bootstrap intervals was 0. 12 IPSS points, well below the minimal clinically important difference, across all learners.

The conclusion survives the learner panel and the transposition onto a binary target. p. 8
Where the paper says it

66, driven by folds in which the regularised model is destabilised by the added dimensions, but these are degradations and do not weaken the ceiling on benefit. The conclusion survives the learner panel and the transposition onto a binary target.

Discussion on Methodology

The concordance analysis is a procedure rather than a verdict on two particular neural networks, as the DVH index margin encodes the acceptable dosimetric shift derived from human contouring IOV. The patient-level threshold tells what the cohort test cannot, and is itself not a constant. p. 9

Where the paper says it

A cohort verdict bears on a mean, and here the individual dispersion exceeds the margin by an order of magnitude while the mean consumes a fraction of it, so equivalence on average carries no implication for the patient in front of the physician. The patient-level threshold tells what the cohort test cannot, and is itself not a constant: on 𝐷90 and 𝑉100 a fixed contouring error costs more overlap on a small gland, so the quality required for reliability rises as the prostate shrinks, whereas on the two high-dose volumes the dependence is additive and runs the other way.

The null result on the DVH indices predictive power is informative only if the clinical baseline M0 left something to gain. p. 9
Where the paper says it

On the predictive side the dose-volume panel adds nothing to a clinical baseline for ΔIPSS, and the segmentation source is irrelevant to that absence. This null result is bounded and robust: at every horizon the upper end of the confidence interval IPSS gain sits well below the MCID, and the corrected 𝑡-tests and the agreement across learner models is telling.

Learner Panel Analysis

Across the twenty-four tests, four DVH blocks at six horizons with the pre-specified ElasticNet, no cell satisfied the declared criteria for significance. The controls behave as designed and confirm that the apparatus detects what it should. p. 8

Where the paper says it

79, both detected at nearly every horizon, while the negative controls produce no detected cell. The binary target recovers the total IPSS ablation but detects the obstructive subscore no more often than age or prostate volume, so its detection floor lies above the largest DVH shift observed and it corroborates the continuous analysis rather than standing on its own (Supplementary S7).

Ablating the pre-treatment total IPSS costs between 0. p. 8
Where the paper says it

05, and no DVH block shifts the AUC of the binary target by more than +0. The controls calibrate what each target can detect: ablating the pre-treatment total IPSS costs +0.

38 to 1.46 IPSS points depending on the horizon, with six intervals out of six excluding zero and six corrected tests out of six significant for the ElasticNet learner. p. 8
Where the paper says it

The conclusion survives the learner panel and the transposition onto a binary target. Over five algorithms and six horizons the DVH blocks yield at most three bootstrap intervals excluding zero out of six, without sign consistency and never with a corrected 𝑝-value below 0.

The obstructive subscore alone gives +0. p. 4
Where the paper says it

05, and reproduction of sign and order of magnitude across the learner panel. Positive controls ablate the pre-treatment IPSS, full score or obstructive subscore alone, negative controls age and prostate volume.

23 to +0.79, detected at five horizons out of six by both criteria. p. 8
Where the paper says it

The conclusion survives the learner panel and the transposition onto a binary target. Over five algorithms and six horizons the DVH blocks yield at most three bootstrap intervals excluding zero out of six, without sign consistency and never with a corrected 𝑝-value below 0.

Baseline Performance and Headroom

The baseline retains headroom at every horizon, with the tightest case being 4 years. Under the most conservative floor, the baseline still carries at least 1. p. 4

Where the paper says it

The primary inferential object of the outcome analysis is the per-patient difference in root mean squared error between paired conditions, ΔRMSE𝑖 = RMSE(𝑀0 ) 𝑖 − RMSE(𝑀1 ) 𝑖 , positive values indicating improvement, with a percentile bootstrap 95 % confidence interval over 5000 replicates in which patients are the resampling unit so that pairing is preserved. It is complemented by a corrected Nadeau-Bengio resampled 𝑡-test[35] on the per-fold differences, the naive paired 𝑡-test being anti-conservative because training sets overlap across folds[5].

17 IPSS points of removable error at every horizon. p. 8
Where the paper says it

97 IPSS points. 43 points above the measurement floor of the instrument at every horizon, so a predictor carrying meaningful information would have had room to demonstrate it (Supplementary S6).

The largest shift produced by any DVH indices block is 0. p. 8
Where the paper says it

79, both detected at nearly every horizon, while the negative controls produce no detected cell. The binary target recovers the total IPSS ablation but detects the obstructive subscore no more often than age or prostate volume, so its detection floor lies above the largest DVH shift observed and it corroborates the continuous analysis rather than standing on its own (Supplementary S7).

055 points and the median over the twenty-four tests (four DVH blocks at six horizons) is 0. p. 8
Where the paper says it

43 points above the measurement floor of the instrument at every horizon, so a predictor carrying meaningful information would have had room to demonstrate it (Supplementary S6). Across the twenty-four tests, four DVH blocks (one per segmentation source + Bayesian variance) at six horizons with the pre-specified ElasticNet, no cell satisfied the declared criteria for significance (Fig.

02, so the reported effects are at least twenty times smaller than the room available. p. 8
Where the paper says it

43 points above the measurement floor of the instrument at every horizon, so a predictor carrying meaningful information would have had room to demonstrate it (Supplementary S6). Across the twenty-four tests, four DVH blocks (one per segmentation source + Bayesian variance) at six horizons with the pre-specified ElasticNet, no cell satisfied the declared criteria for significance (Fig.

Binary Target Performance

The dedicated logistic regression is the only row in which a DVH block reaches declared significance, and it does so with the wrong sign, −0.048 for the manual block. p. 8

Where the paper says it

12 (Manual DVH) 0. 04 (Deterministic auto DVH) IPSS MCID 3 points (b) Figure 4: Incremental value of the dose-volume block over the clinical baseline for the prediction of ΔIPSS with ElasticNet learner.

The detection floor of the binary target therefore lies between roughly 0. p. 8
Where the paper says it

79, both detected at nearly every horizon, while the negative controls produce no detected cell. The binary target recovers the total IPSS ablation but detects the obstructive subscore no more often than age or prostate volume, so its detection floor lies above the largest DVH shift observed and it corroborates the continuous analysis rather than standing on its own (Supplementary S7).

04 and 0.08 of area under the curve, well above the largest shift any DVH block produces here, +0. p. 8
Where the paper says it

79, both detected at nearly every horizon, while the negative controls produce no detected cell. The binary target recovers the total IPSS ablation but detects the obstructive subscore no more often than age or prostate volume, so its detection floor lies above the largest DVH shift observed and it corroborates the continuous analysis rather than standing on its own (Supplementary S7).

020.

Conclusion on Predictive Power

Deep learning contours reproduce the expert dose-volume indices closely enough that the choice of source does not change what those indices predict, which in this cohort is nothing beyond the clinical baseline. The reliability question and the predictive question are separable, and both are answerable against margins and criteria declared in advance rather than read off the data. p. 9

Where the paper says it

Deep learning contours reproduce the expert dose-volume indices closely enough that the choice of source does not change what those indices predict, which in this cohort is nothing beyond the clinical baseline. The reliability question and the predictive question are separable, and both are answerable against margins and criteria declared in advance rather than read off the data.

Data Availability

The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy considerations and institutional ethics restrictions. De-identified data may be made available from the corresponding author upon reasonable request, subject to approval by relevant institutional research ethics boards and data-sharing agreements where applicable.

Generative AI Usage

During the preparation of this work the authors used Claude (Anthropic) to assist with the development of the analysis code and core text length reduction. The first author reviewed and validated all generated code and takes full responsibility for the content of the published article. p. 4

Where the paper says it

Positive controls ablate the pre-treatment IPSS, full score or obstructive subscore alone, negative controls age and prostate volume. Analysis code is publicly available (https://github.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Adaptive equivalence margin setting: The system can dynamically adjust its acceptance threshold based on prostate volume to define a patient-level quality metric, as described by The clinically recorded prostate volume was then added, and the threshold model chosen between an additive form, Dice with volume, and one carrying their interaction.

  2. Volume-dependent quality scoring: The system can use a logistic regression model to determine the segmentation quality score as a predictor of equivalence at 90 percent probability. This score is derived by comparing "eight candidate metrics spanning overlap, distance and volumetric families as predictors of E (E = 1 Delta <= delta)."

  3. Directional error analysis: The system can calculate the signed index difference regressed on the signed relative volume error to determine in which direction and by how much it departs from its manual counterpart when it is not, providing a metric orthogonal to unsigned metrics like Dice.

  4. Robust predictive null hypothesis testing: The system can utilize a corrected t-test, incorporating the covariance induced by cross-validation overlap, to detect incremental value, as this test is a condition of admissibility for declaring significance in outcome analysis.

  5. Uncertainty propagation management: The system can leverage Bayesian neural networks with variational conversion and posterior sampling to generate predictive uncertainty estimates that quantify the model's uncertainty about its own weights rather than just its distance from an expert contour.

Abstract

Background and purpose: Dose Volume Histogram (DVH) indices remain the main dose-effect metrics for toxicity prediction, but they depend on how contours were made. Inter-observer variability (IOV) is unavoidable and clinically accepted, so the question for automatic segmentation addresses equivalence: do automatic contours produce DVH errors comparable to human IOV, and can these indices predict patient-reported toxicity? Materials and methods: In 429 patients treated with iodine-125 LDR-BT monotherapy, indices from expert manual delineation were compared with a deterministic and a Bayesian nnUNet on a fixed dose distribution, by two-one-sided tests against margins set from CT contouring IOV. Logistic regression gave patient-level thresholds at 90,% probability of equivalence. In 380 patients, eleven DHV indices were added to a clinical baseline predicting change in International Prostate Symptom Score (IPSS) at six horizons over 5 years, with nested cross-validation, bootstrap intervals and corrected t-tests. Results: All cohort-level comparisons of DVH indices were declared equivalent, with no interval consuming more than 44,% of its equivalence margin. Individual agreement was weaker with equivalence rates from 55.9,% to 89.7,%, depending on the DVH index and automatic segmentation model. Thresholds ranged from 0.864 to 0.958 Dice. No DVH block improved IPSS prediction at any horizon. The largest improvement declared by bootstrap intervals was 0.12 IPSS points, well below the minimal clinically important difference, across all learners. Conclusion: Automatic contours matched expert dosimetry within human IOV on average, but individual equivalence requires a volume dependent quality metric threshold definition. Our DVH indices panel provides no significant predictive power for IPSS, irrespective of segmentation source and horizon.

Related papers