Prostate Cancer Screening with Artificial Intelligence-Enhanced Micro-Ultrasound: A Comparative Study with Traditional Methods

Jason P. Joseph; Li-Ming Su; Muhammad Imran; Wayne G. Brisbane; Wei Shao

REVIEW 5 major objections 4 minor 15 references

Prostate Cancer Screening with Artificial Intelligence-Enhanced Micro-Ultrasound: A Comparative Study with Traditional Methods

T0 review · 5 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read AI reading of micro-ultrasound detects clinically significant prostate cancer more accurately than PSA and DRE.

desk verdict A genuinely new micro-US slice-level alignment for prostate cancer screening, but the headline AUROC is not reproducible from the methods as written until the continuous patient-level score and threshold tuning are clarified. read the letter →

arxiv 2505.21355 v1 pith:TCNFM7XP submitted 2025-05-27 eess.IV cs.AIcs.CV

Muhammad Imran , Wayne G. Brisbane , Li-Ming Su , Jason P. Joseph , Wei Shao This is my paper

classification eess.IVcs.AIcs.CV

keywords prostatecancerscreeningmicro-ultrasoundclinicallysignificantartificialintelligenceself-supervisedlearningconvolutionalautoencoderrandomforestPSA

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that artificial intelligence can interpret micro-ultrasound images of the prostate well enough to outperform standard clinical screening based on PSA, DRE, age, and prostate volume. In a retrospective cohort of 145 men who underwent micro-ultrasound-guided biopsy (79 with clinically significant prostate cancer, Gleason 3+4 or higher, and 66 without), the AI imaging model reached a patient-level AUROC of 0.871 versus 0.753 for the clinical model. At a fixed threshold it kept sensitivity at 92.5% while raising specificity to 68.1%, compared with 96.2% sensitivity and 27.3% specificity for the clinical model. If this holds outside the training cohort, it would mean a low-cost, point-of-care imaging test that spares many men unnecessary biopsies without sacrificing detection of aggressive disease.

What carries the argument

The load-bearing mechanism is a two-stage pipeline: a self-supervised convolutional autoencoder trained to reconstruct 2D micro-US slices learns a 256-dimensional feature representation for each slice, then a random forest classifies slices as containing clinically significant cancer or not. Patient-level predictions are derived by a spatial contiguity rule: a patient is positive if at least eight consecutive slices are predicted positive, a threshold chosen from retrospective lesion-length analysis to reduce false positives. The comparison arm is a random forest trained on the standard clinical variables PSA, DRE, age, and prostate volume. Slice-level training labels come from an expert urologist mapping biopsy needle trajectories onto pre-biopsy scans using PRI-MUS sonographic features.

What would settle it

Take the trained model to an independent multi-center cohort with whole-mount histology registered to micro-US slices and compare its patient-level calls with a panel of blinded expert readers at the same 0.15 threshold. If the AUROC no longer exceeds the clinical model's 0.753, or the 92.5% sensitivity and 68.1% specificity point is not reproduced, the claim is refuted.

Watch

Extended reading notes

Core claim

The central claim is that AI-interpreted micro-ultrasound is a stronger screening test for clinically significant prostate cancer than traditional biomarkers. Using a self-supervised convolutional autoencoder as a fixed feature extractor and a random forest slice-level classifier, with patients called positive when at least eight consecutive slices are predicted positive, the imaging model achieved AUROC 0.871. The clinical model built on PSA, DRE, prostate volume, and age achieved AUROC 0.753. At a 0.15 probability threshold the micro-US model achieved 92.5% sensitivity and 68.1% specificity, while the clinical model achieved 96.2% sensitivity and 27.3% specificity; the imaging model also had higher accuracy, precision, and F1-score. The authors interpret this as evidence that AI can provide the specificity PSA lacks while keeping sensitivity high, potentially reducing unnecessary biopsies.

Load-bearing premise

The comparison rests on slice-level cancer labels made by one expert urologist mapping biopsy needle trajectories onto pre-biopsy micro-ultrasound images without independent verification; if those labels are inaccurate or idiosyncratic, the reported AUROC and specificity gains will not reproduce.

Editorial extensions

If this is right

If the AUROC gap replicates prospectively, AI-interpreted micro-US could serve as a first-line screening test that refers far fewer men for biopsy than PSA-based screening at the same detection rate.
At 68.1% specificity versus 27.3%, roughly two of every three men without clinically significant cancer would be spared an unnecessary biopsy that PSA-based screening would trigger.
Because micro-US is portable and inexpensive, an AI reader could extend accurate prostate cancer screening beyond MRI-equipped centers.
The eight-consecutive-slice rule can be tuned to shift the operating point toward higher sensitivity or higher specificity for different clinical settings.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper evaluates imaging and clinical variables separately; combining the AI features with PSA, DRE, age, and volume in one model could push AUROC above 0.871, a testable next step.
Because the slice-level labels come from a single expert's cognitive mapping of biopsy trajectories, the reported performance may partially reflect that expert's skill; comparing the same pipeline trained on labels from multiple readers or whole-mount histology would isolate the AI's own contribution.
The cohort is biopsy-referred men with elevated PSA or abnormal DRE, so the 68.1% specificity is measured in that enriched population; a screening cohort with lower disease prevalence would likely show different operating-point behavior and should be tested before adoption.
The contiguity rule suggests a link between lesion length and detectability; the same architecture could be evaluated for detecting cancer size thresholds beyond Gleason grade, such as lesion volume.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Two things you should know about arXiv:2505.21355. First, it is genuinely the first slice-level alignment of micro-US images with biopsy pathology for patient-level screening, and the comparison against a clinical biomarker model (PSA, DRE, age, volume) is a real new result. Second, the headline AUROC gap (0.871 vs 0.753) is not reproducible from the methods as written, and I don't think the superiority claim is established yet.

What's good: the cohort is real (145 men, 79 csPCa), the autoencoder/random forest pipeline is described in enough detail to be replicated, and the authors are upfront about single-center retrospective design, lack of external validation, and the empirically chosen 8-consecutive-slices rule. They also compare against a reasonable clinical baseline, not just chance.

Where it gets soft: the patient-level AUROC is the primary evidence, but Section 2.2.2 gives only a binary aggregation rule (≥8 consecutive positive slices) and Section 2.2.6 gives a probability threshold of 0.15. Neither defines the continuous patient score that an ROC curve needs. So the 0.871/0.753 AUROC cannot be audited from what's written. On top of that, both the 8-slice rule and the 0.15 threshold are determined from the same cohort used for evaluation, which biases the threshold-based metrics (92.5% sensitivity/68.1% specificity) in the imaging model's favor; the clinical model has no analogous tuned aggregation. The slice labels come from one expert's cognitive mapping of biopsy trajectories to pre-biopsy slices, with all ambiguous slices in positive patients excluded, so the model may be learning the expert's PRI-MUS impression rather than biopsy-confirmed cancer. No confidence intervals are given, so the 0.118 AUROC gap could be sampling noise.

I'd send this to peer review, not desk reject. The application is clinically meaningful, the data are real, and the core idea (slice-level alignment) is worth refereeing. But it needs major revision: define the patient-level score, use nested cross-validation or a development/test split for threshold selection, include CIs, and ideally bring in a second reader or whole-mount histology to validate the slice labels. If those are addressed, the result might hold. As written, I wouldn't cite it yet.

For your reading group: maybe, mainly as a case study in how threshold tuning on the test set can make an AUROC story look cleaner than it is.

Referee Report

5 major / 4 minor

Summary. The manuscript presents a retrospective, single-center study of 145 men who underwent micro-ultrasound-guided prostate biopsy. The authors train a convolutional autoencoder in a self-supervised manner on 2D micro-US slices to extract 256-dimensional features, then use a random forest to classify slices as indicative of clinically significant prostate cancer. A patient is classified as csPCa-positive if at least eight consecutive slices are predicted positive. The imaging model is compared against a clinical model using PSA, DRE, age, and prostate volume. The reported results are a patient-level AUROC of 0.871 for the imaging model versus 0.753 for the clinical model, and at a fixed threshold of 0.15, 92.5% sensitivity with 68.1% specificity for the imaging model versus 96.2% sensitivity with 27.3% specificity for the clinical model. The authors conclude that AI-interpreted micro-US improves specificity while maintaining high sensitivity for csPCa detection.

Significance. If the reported performance holds under unbiased evaluation, the work would be clinically meaningful: a low-cost, point-of-care AI screening tool that reduces unnecessary biopsies would be valuable. The study has several strengths: it uses a self-supervised autoencoder to avoid relying on large annotated image datasets, attempts slice-level alignment with biopsy cores, and directly compares the imaging model with a clinical biomarker model in the same cohort. The paper also acknowledges its retrospective single-center design and lack of external validation. However, the evaluation methodology contains load-bearing gaps: the patient-level AUROC is not reproducible from the methods, the aggregation rule and decision threshold are tuned on the same evaluation data, and the slice-level ground truth is based on one expert's subjective cognitive mapping. These issues prevent the results, as presented, from substantiating the superiority claim.

major comments (5)

[§2.2.2, §2.2.6] The patient-level AUROC of 0.871 is not defined by the methods as written. Sections 2.2.2 and 2.2.6 define a binary patient-level rule (at least eight consecutive positive slices) and a slice-level probability threshold (0.15), but an AUROC requires a continuous patient-level score to rank patients. Without specifying how the patient-level score is computed (e.g., maximum slice probability, mean probability, or count of positive slices), the headline AUROC cannot be reproduced or audited. Please provide the exact aggregation used to generate the ROC curve and report confidence intervals for the AUROC values.
[§2.2.2, §2.2.6, §4] The patient-level aggregation rule (≥8 consecutive positive slices) and the decision threshold of 0.15 are both derived from the same cohort used for evaluation. Section 2.2.2 states the eight-slice rule is based on a retrospective analysis of lesion length in this cohort, and Section 2.2.6 states the 0.15 threshold was empirically selected to balance sensitivity and specificity in the training data. Because the same data are used to set these parameters and to report the metrics, the sensitivity, specificity, accuracy, and AUROC are optimistically biased. Importantly, the clinical comparator has no analogous tuned aggregation rule, making the comparison unfair. Use nested cross-validation, a separate validation set, or a predefined rule to fix these parameters before evaluation, and report results across a range of thresholds.
[§2.1.3] The slice-level ground truth for csPCa-positive patients is generated by a single expert urologist (WGB) who cognitively maps each biopsy core's trajectory to pre-biopsy micro-US slices and labels suspicious slices as positive; all unlabeled slices in these patients are excluded from training. This protocol means the model may learn the expert's subjective PRI-MUS-based impression rather than an objective, biopsy-confirmed label, and the exclusion of unlabeled slices makes the training set non-representative of the full gland. No independent verification (e.g., whole-mount histology, reader agreement, or a second expert) is provided. Please report the label repeatability or add a sensitivity analysis using alternative labeling strategies, and discuss the potential bias in the conclusions.
[§2.2.4] The cross-validation description is internally inconsistent. The text says the dataset is partitioned into five mutually exclusive folds, but then states that each patient appears exactly once in the test set, once in the validation set, and in the training set for the remaining three folds. In a standard five-fold cross-validation, each patient appears in the test set once and in the training set four times; the described scheme with a separate validation fold would require a different split (e.g., repeated holdout or nested CV). This ambiguity affects the credibility of the reported averaged metrics. Please clarify the exact splitting and hyperparameter selection procedure, and specify how the validation set was used.
[§3.1, §3.2] No confidence intervals or statistical significance tests are reported for the AUROC difference (0.871 vs. 0.753) or for the threshold-based metrics. With 145 patients and a 0.118 AUROC gap, the difference may be within sampling variability. Report bootstrapped confidence intervals for all metrics and a DeLong test (or equivalent) for the AUROC comparison, and report patient-level, not slice-level, uncertainty.

minor comments (4)

[Abstract] In the abstract, 'if 88 or more consecutive slices were predicted positive' should read 'if ≥8 consecutive slices were predicted positive' (the digit '8' appears as '88').
[§2.2.6] The section heading contains a typo: 'Peformance metrics' should be 'Performance metrics'.
[§2.2.3] The sentence 'Since the scikit-learn implementation supports internal out-of-bag (OOB) validation, a separate validation set was not required for hyperparameter tuning' is unclear. OOB scores estimate generalization error and are not a substitute for explicit hyperparameter tuning; please describe how the random forest hyperparameters were actually selected.
[Figure 2] The ROC curves should specify whether they are pooled across all test-fold predictions or averaged over folds, and if averaged, error bars or shaded intervals should be shown.

Circularity Check

2 steps flagged · score 6.0 of 10

Patient-level AUROC and threshold metrics are partly tuned on the evaluated cohort; the slice-level classifier itself is cross-validated, so the circularity is partial, not total.

fitted input called prediction [Section 2.2.6 (Performance metrics); Table 2]
"The remaining metrics were computed using a fixed probability threshold of 0.15, which was empirically selected to balance sensitivity and specificity in the training data."

The reported sensitivity (92.5%) and specificity (68.1%) are evaluated at exactly the threshold that was chosen by optimizing the sensitivity/specificity trade-off on the data. As written, the threshold is a cohort-fitted operating point rather than a prespecified or externally derived cut-point; the same data are therefore used both to select the decision rule and to report its performance. This does not force the imaging-versus-clinical AUROC difference, but it makes the headline operating-point metrics partly in-sample by construction.
fitted input called prediction [Section 2.2.2 (Micro-US image classification); Section 4 (Discussion)]
"A patient was considered csPCa-positive if at least eight consecutive slices were predicted positive. This rule was based on retrospective analysis of lesion length, which showed that csPCa typically spanned eight adjacent slices on average."

The patient-level aggregation rule is derived from a retrospective analysis of the same cohort on which patient-level AUROC is then reported. The Discussion concedes that this rule 'was empirically defined and may require adjustment,' confirming it is a fitted hyperparameter rather than a first-principles or prespecified rule. Because the same cohort is used both to fit the aggregation rule and to evaluate the resulting patient-level predictions, the reported patient-level discrimination is not an independent test of a fixed decision rule. The slice-level random forest is cross-validated, so this is partial circularity rather than a fully forced result.

full rationale

The paper's central chain—micro-US slices, autoencoder features, slice-level random forest, patient-level aggregation, AUROC—is an empirical evaluation rather than a formal derivation from first principles. The most significant circularity burden is that two decision parameters are empirically fitted to the study cohort and then used to report the headline patient-level metrics: the 0.15 probability threshold (Section 2.2.6) and the 'at least eight consecutive slices' rule (Section 2.2.2). Both are acknowledged as empirically chosen, and the threshold is selected to balance sensitivity and specificity on the same data whose reporting subsequently uses that threshold. The slice-level classifier itself is trained with five-fold cross-validation, and the comparison against a clinical biomarker model has independent content, so the work is not wholly circular. However, as written, the patient-level score used for the AUROC is never defined, no confidence intervals are reported, and the paper does not specify that threshold and aggregation-rule selection were nested inside the cross-validation folds. These are reproducibility and statistical-validity concerns rather than circularity per se, but they compound the fitted-parameter issue. The presence of cohort-tuned aggregation with no external validation places the paper at partial circularity: score 6.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The main free parameters are the two thresholds used to turn slice-level model outputs into patient-level screening decisions. Both were derived from the same 145-patient cohort on which the model is evaluated, so the reported performance partially reflects fitting to the evaluation data. No new physical entities or unobserved constructs are introduced.

free parameters (3)

Probability threshold = 0.15
Empirically selected on training data to balance sensitivity and specificity; the reported sensitivity, specificity, accuracy, precision, and F1-score all depend on this cutoff. Section 2.2.6.
Minimum consecutive positive slices for patient-level positivity = 8
Chosen after retrospective analysis of lesion length in the same cohort; this rule converts slice-level predictions into patient-level outcomes and therefore affects the patient-level AUROC as well as the other metrics. Section 2.2.2.
Autoencoder latent dimension = 256
Design choice for feature extractor capacity; it shapes the features used by the random forest but is not fitted to the cancer label. Included for completeness.

assumptions (4)

domain assumption Clinically significant prostate cancer is defined by any biopsy core with Gleason score >= 3+4.
Standard clinical definition, but biopsy-based ground truth may miss cancers present in un-sampled regions.
domain assumption PRI-MUS sonographic features are a valid basis for labeling suspicious micro-US slices.
The expert used PRI-MUS criteria to mark positive slices; if PRI-MUS is not reliably tied to csPCa, the labels are noisy. Section 2.1.3.
domain assumption The cognitive mapping of biopsy cores to pre-biopsy ultrasound slices is accurate.
The correspondence between in-vivo biopsy location and pre-biopsy image is approximate; errors would mislabel slices and bias training.
standard math Random forest and autoencoder training converge to stable solutions with the given hyperparameters.
No formal guarantees; relies on standard machine-learning practice.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Prostate Cancer Screening with Artificial Intelligence-Enhanced Micro-Ultrasound: A Comparative Study with Traditional Methods." pith.science (2026). https://pith.science/paper/TCNFM7XP

@misc{pith2026250521355,
  author       = {Pith},
  title        = {Pith review of: Prostate Cancer Screening with Artificial Intelligence-Enhanced Micro-Ultrasound: A Comparative Study with Traditional Methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TCNFM7XP}},
  note         = {Machine review of arXiv:2505.21355}
}

read the original abstract

Background and objective: Micro-ultrasound (micro-US) is a novel imaging modality with diagnostic accuracy comparable to MRI for detecting clinically significant prostate cancer (csPCa). We investigated whether artificial intelligence (AI) interpretation of micro-US can outperform clinical screening methods using PSA and digital rectal examination (DRE). Methods: We retrospectively studied 145 men who underwent micro-US guided biopsy (79 with csPCa, 66 without). A self-supervised convolutional autoencoder was used to extract deep image features from 2D micro-US slices. Random forest classifiers were trained using five-fold cross-validation to predict csPCa at the slice level. Patients were classified as csPCa-positive if 88 or more consecutive slices were predicted positive. Model performance was compared with a classifier using PSA, DRE, prostate volume, and age. Key findings and limitations: The AI-based micro-US model and clinical screening model achieved AUROCs of 0.871 and 0.753, respectively. At a fixed threshold, the micro-US model achieved 92.5% sensitivity and 68.1% specificity, while the clinical model showed 96.2% sensitivity but only 27.3% specificity. Limitations include a retrospective single-center design and lack of external validation. Conclusions and clinical implications: AI-interpreted micro-US improves specificity while maintaining high sensitivity for csPCa detection. This method may reduce unnecessary biopsies and serve as a low-cost alternative to PSA-based screening. Patient summary: We developed an AI system to analyze prostate micro-ultrasound images. It outperformed PSA and DRE in detecting aggressive cancer and may help avoid unnecessary biopsies.

Figures

Figures reproduced from arXiv: 2505.21355 by the authors.

**Figure 2.** ROC curves comparing the imaging-based model and clin [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

15 extracted references · 14 canonical work pages

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...
[2]

, author Bosaily, A.E.S

author Ahmed, H.U. , author Bosaily, A.E.S. , author Brown, L.C. , author Gabe, R. , author Kaplan, R. , author Parmar, M.K. , author Collaco-Moraes, Y. , author Ward, K. , author Hindley, R.G. , author Freeman, A. , et al., year 2017 . title Diagnostic accuracy of multi-parametric mri and trus biopsy in prostate cancer (promis): a paired validating confi...

work page 2017
[3]

, author Wilson, P

author Gilany, M. , author Wilson, P. , author Perera-Ortega, A. , author Jamzad, A. , author To, M.N.N. , author Fooladgar, F. , author Wodlinger, B. , author Abolmaesumi, P. , author Mousavi, P. , year 2023 . title Trusformer: improving prostate cancer detection from micro-ultrasound using attention and self-supervision . journal International Journal o...

work page 2023
[4]

, author Wilson, P.F

author Harmanani, M. , author Wilson, P.F. , author To, M.N.N. , author Gilany, M. , author Jamzad, A. , author Fooladgar, F. , author Wodlinger, B. , author Abolmaesumi, P. , author Mousavi, P. , year 2025 . title Trusworthy: toward clinically applicable deep learning for confident detection of prostate cancer in micro-ultrasound . journal International ...

work page 2025
[5]

, author Nguyen, B

author Imran, M. , author Nguyen, B. , author Pensa, J. , author Falzarano, S.M. , author Sisk, A.E. , author Liang, M. , author DiBianco, J.M. , author Su, L.M. , author Zhou, Y. , author Joseph, J.P. , et al., year 2024 . title Image registration of in vivo micro-ultrasound and ex vivo pseudo-whole mount histopathology images of the prostate: A proof-of...

work page 2024
[6]

, author Tannock, I

author James, N.D. , author Tannock, I. , author N'Dow, J. , author Feng, F. , author Gillessen, S. , author Ali, S.A. , author Trujillo, B. , author Al-Lazikani, B. , author Attard, G. , author Bray, F. , et al., year 2024 . title The lancet commission on prostate cancer: planning for the surge in cases . journal The Lancet volume 403 , pages 1683--1722

work page 2024
[7]

, author Imran, M

author Jiang, H. , author Imran, M. , author Muralidharan, P. , author Patel, A. , author Pensa, J. , author Liang, M. , author Benidir, T. , author Grajo, J.R. , author Joseph, J.P. , author Terry, R. , et al., year 2024 . title Microsegnet: A deep learning approach for prostate segmentation on micro-ultrasound images . journal Computerized Medical Imagi...

work page 2024
[8]

, author Luger, F

author Kinnaird, A. , author Luger, F. , author Cash, H. , author Ghai, S. , author Urdaneta-Salegui, L.F. , author Pavlovich, C.P. , author Brito, J. , author Shore, N.D. , author Struck, J.P. , author Schostak, M. , et al., year 2025 . title Microultrasonography-guided vs mri-guided biopsy for prostate cancer diagnosis: The optimum randomized clinical t...

work page 2025

Show all 15 references

[9]

, author Lughezzani, G

author Klotz, L. , author Lughezzani, G. , author Maffei, D. , author S \'a nchez, A. , author Pereira, J.G. , author Staerman, F. , author Cash, H. , author Luger, F. , author Lopez, L. , author Sanchez-Salas, R. , et al., year 2020 . title Comparison of micro-ultrasound and ...

2020
[10]

, author Saita, A

author Lughezzani, G. , author Saita, A. , author Lazzeri, M. , author Paciotti, M. , author Maffei, D. , author Lista, G. , author Hurle, R. , author Buffi, N.M. , author Guazzoni, G. , author Casale, P. , year 2019 . title Comparison of the diagnostic accuracy of micro-ultra...

2019
[11]

, author Avolio, P.P

author Maffei, D. , author Avolio, P.P. , author Moretto, S. , author Piccolini, A. , author Aljoulani, M. , author Dagnino, F. , author De Carne, F. , author Fasulo, V. , author Marco, P. , author Saita, A.R. , et al., year 2024 . title Mp49-15 evaluating the role of pri-mus ...

2024
[12]

, author Kruger, A.B

author Schr \"o der, F.H. , author Kruger, A.B. , author Rietbergen, J. , author Kranse, R. , author Maas, P.v.d. , author Beemsterboer, P. , author Hoedemaeker, R. , year 1998 . title Evaluation of the digital rectal examination as a screening test for prostate cancer . journ...

1998
[13]

, year 2025

author Society, A.C. , year 2025 . title Cancer facts & figures 2025. atlanta: American cancer society; 2025

2025
[14]

, author Pauler, D.K

author Thompson, I.M. , author Pauler, D.K. , author Goodman, P.J. , author Tangen, C.M. , author Lucia, M.S. , author Parnes, H.L. , author Minasian, L.M. , author Ford, L.G. , author Lippman, S.M. , author Crawford, E.D. , et al., year 2004 . title Prevalence of prostate can...

2004
[15]

, author Choi, M.H

author Zhou, S.R. , author Choi, M.H. , author Vesal, S. , author Kinnaird, A. , author Brisbane, W.G. , author Lughezzani, G. , author Maffei, D. , author Fasulo, V. , author Albers, P. , author Zhang, L. , et al., year 2024 . title Inter-reader agreement for prostate cancer ...

2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

[2] [2]

, author Bosaily, A.E.S

author Ahmed, H.U. , author Bosaily, A.E.S. , author Brown, L.C. , author Gabe, R. , author Kaplan, R. , author Parmar, M.K. , author Collaco-Moraes, Y. , author Ward, K. , author Hindley, R.G. , author Freeman, A. , et al., year 2017 . title Diagnostic accuracy of multi-parametric mri and trus biopsy in prostate cancer (promis): a paired validating confi...

work page 2017

[3] [3]

, author Wilson, P

author Gilany, M. , author Wilson, P. , author Perera-Ortega, A. , author Jamzad, A. , author To, M.N.N. , author Fooladgar, F. , author Wodlinger, B. , author Abolmaesumi, P. , author Mousavi, P. , year 2023 . title Trusformer: improving prostate cancer detection from micro-ultrasound using attention and self-supervision . journal International Journal o...

work page 2023

[4] [4]

, author Wilson, P.F

author Harmanani, M. , author Wilson, P.F. , author To, M.N.N. , author Gilany, M. , author Jamzad, A. , author Fooladgar, F. , author Wodlinger, B. , author Abolmaesumi, P. , author Mousavi, P. , year 2025 . title Trusworthy: toward clinically applicable deep learning for confident detection of prostate cancer in micro-ultrasound . journal International ...

work page 2025

[5] [5]

, author Nguyen, B

author Imran, M. , author Nguyen, B. , author Pensa, J. , author Falzarano, S.M. , author Sisk, A.E. , author Liang, M. , author DiBianco, J.M. , author Su, L.M. , author Zhou, Y. , author Joseph, J.P. , et al., year 2024 . title Image registration of in vivo micro-ultrasound and ex vivo pseudo-whole mount histopathology images of the prostate: A proof-of...

work page 2024

[6] [6]

, author Tannock, I

author James, N.D. , author Tannock, I. , author N'Dow, J. , author Feng, F. , author Gillessen, S. , author Ali, S.A. , author Trujillo, B. , author Al-Lazikani, B. , author Attard, G. , author Bray, F. , et al., year 2024 . title The lancet commission on prostate cancer: planning for the surge in cases . journal The Lancet volume 403 , pages 1683--1722

work page 2024

[7] [7]

, author Imran, M

author Jiang, H. , author Imran, M. , author Muralidharan, P. , author Patel, A. , author Pensa, J. , author Liang, M. , author Benidir, T. , author Grajo, J.R. , author Joseph, J.P. , author Terry, R. , et al., year 2024 . title Microsegnet: A deep learning approach for prostate segmentation on micro-ultrasound images . journal Computerized Medical Imagi...

work page 2024

[8] [8]

, author Luger, F

author Kinnaird, A. , author Luger, F. , author Cash, H. , author Ghai, S. , author Urdaneta-Salegui, L.F. , author Pavlovich, C.P. , author Brito, J. , author Shore, N.D. , author Struck, J.P. , author Schostak, M. , et al., year 2025 . title Microultrasonography-guided vs mri-guided biopsy for prostate cancer diagnosis: The optimum randomized clinical t...

work page 2025