The Algorithm in the Lab: AI is changing the pathologist job
An Automated, Pathologist-free Gleason Grade Stratifies Disease-free Interval Comparably to Expert Grading from a Single Out-of-distribution Slide | medRxiv
How Artificial Intelligence Is Changing the Way Your Biopsy Is Read
AI tools that grade prostate cancer as well as expert pathologists are winning FDA approval, entering clinical guidelines, and moving toward routine use — and a new preprint study shows they can now do it without a pathologist in the room at all.
If you have been diagnosed with prostate cancer, the single most important number in your medical record — your Gleason grade — was almost certainly assigned by a pathologist staring through a microscope (or these days, a computer screen) at thin slices of your biopsy tissue. That expert human judgment has guided treatment decisions for decades. Now, a wave of artificial intelligence (AI) tools is reaching the clinic that can perform the same task — automatically, consistently, and at scale — and the results are challenging some long-held assumptions about how that grading process works, and who (or what) should do it.
This article walks you through what these AI systems do, how good they actually are, what the FDA has approved, and what a striking new research study — posted to the medical preprint server medRxiv in June 2026 — tells us about the future of pathology-free cancer grading.
Why Gleason Grading Matters — and Where It Falls Short
The Gleason grading system, updated by the International Society of Urological Pathology (ISUP) in 2014 into five numbered grade groups (GG1 through GG5), is the single most powerful predictor of how your prostate cancer is likely to behave. GG1 tumors are slow-growing and are often candidates for active surveillance; GG5 tumors are aggressive and usually require prompt, intensive treatment. The grade group directly influences decisions about surgery, radiation, hormone therapy, and watchful waiting. Get it wrong, and patients may be overtreated or undertreated.
The problem is that Gleason grading is a subjective judgment. Even experienced urological pathologists looking at the same slide can disagree. Studies have measured this "inter-observer variability" at moderate-to-substantial levels — a quadratic-weighted kappa of roughly 0.56 to 0.70 among specialists, falling to about 0.44 among general pathologists who do not specialize in urological tumors. In practical terms, a patient's grade can differ depending on who reads the slide, at which institution, and on which day. Those differences translate directly into different treatment recommendations.
Compounding this, the number of prostate cancer cases diagnosed each year is rising, while pathology workforce growth is not keeping pace. A global analysis projects that prostate cancer incidence will roughly double by 2040. In the United Kingdom, the Royal College of Pathology reported that only about 3% of departments have adequate staffing, with vacancy rates filled by expensive temporary locum pathologists. In the United States, the demand for pathology services similarly outpaces the available workforce. AI offers a potential answer to both the consistency problem and the capacity problem.
— Dr. Adam Cole, Founder and CSO, TruCore Pathology Group, commenting on the ArteraAI FDA authorization, August 2025
What AI Tools Actually Do
Modern AI pathology systems use a branch of machine learning called "deep learning" — specifically, convolutional neural networks — trained on tens of thousands of digitized biopsy slides, each annotated by expert pathologists. Once trained, the network can analyze a whole-slide image (a gigapixel-resolution digital scan of a glass biopsy slide) and identify, classify, and grade tumor tissue — often in seconds. The best systems work at the level of individual glands, assigning Gleason patterns 3, 4, or 5 to each gland and then computing an overall grade group from the most prevalent patterns, exactly as a pathologist would.
Beyond raw detection and grading, some AI systems now offer additional capabilities that go further than traditional pathology reporting: measuring the precise fraction of high-grade tissue (Gleason 4 and 5), detecting perineural invasion (cancer cells invading nerve sheaths — a marker of more aggressive disease), identifying cribriform architecture (a pattern associated with worse outcomes), quantifying tumor volume, and even predicting long-term outcomes such as the 10-year risk of distant metastasis or cancer-specific death.
The Race to FDA Approval
Three AI pathology tools targeting prostate cancer have now received regulatory clearance from the U.S. Food and Drug Administration — each taking a different approach.
Paige Prostate (Paige AI, New York): The Pioneer
In September 2021, Paige became the first company to receive FDA De Novo marketing authorization for an AI-based pathology product — in any cancer type. Paige Prostate is a cancer detection system: it analyzes digitized biopsy slides, identifies tissue suspicious for cancer, and alerts the pathologist. It does not replace the pathologist's final call; it acts as a second set of eyes, highlighting areas that might have been missed.
The FDA authorization was based on a study in which 16 pathologists examined 527 biopsy slide images — 171 cancerous, 356 benign — with and without the AI. With Paige Prostate, the pathologists' cancer detection rate improved by an average of 7.3% (from 89.5% to 96.8%). More striking, false-negative diagnoses — cases where cancer was missed entirely — dropped by 70%, and false positives dropped by 24%. The AI also helped close the gap between specialist uropathologists and general pathologists, boosting the general pathologists' accuracy to match that of specialists reading without AI assistance.
Ibex Prostate Detect (Ibex Medical Analytics, Boston): The Safety Net
On February 10, 2025, Ibex Medical Analytics received FDA 510(k) clearance for Ibex Prostate Detect (formerly marketed as "Galen Second Read"), making it the second FDA-cleared AI product in digital pathology and the first to receive 510(k) clearance rather than the more burdensome De Novo pathway. The Ibex system focuses on a specific use case: catching cancers that a pathologist initially diagnosed as benign — a "safety net" function applied retrospectively after the initial read.
The Ibex system generates a color-coded heat map overlaid on the whole-slide image, directing attention to areas likely to contain cancer. In validation studies submitted to the FDA, pathologists who read slides without AI assistance missed 13% of cancer cases that the AI subsequently identified. The system achieved a 99.6% positive predictive value for its heat-map accuracy. In a clinical validation study published in Lancet Digital Health, Ibex achieved an area under the receiver-operating characteristic curve (AUC) of 0.991 for cancer detection, 0.941 for distinguishing low-grade from high-grade tumors, and 0.971 for detecting Gleason pattern 5 specifically.
ArteraAI Prostate (Artera, San Francisco): The Prognosticator
The most recent — and arguably the most clinically transformative — FDA action came on August 13, 2025, when Artera received De Novo authorization for ArteraAI Prostate, the first AI software authorized specifically to prognosticate long-term outcomes in patients with non-metastatic prostate cancer. This is a qualitatively different capability from detection or grading: the Artera system does not simply read a grade from the slide; it combines the digital pathology image with the patient's clinical data (a "multimodal AI," or MMAI, approach) to produce individualized estimates of 10-year risk of distant metastasis and prostate-cancer-specific mortality.
The FDA's authorization followed an earlier Breakthrough Device Designation in July 2025 — a special designation reserved for technologies that may provide more effective treatment than existing alternatives — and established an entirely new product code category for AI-powered digital pathology risk-stratification tools. It includes a Predetermined Change Control Plan that allows Artera to expand the platform's capabilities and validate compatibility with new scanners without filing additional 510(k) submissions for each update — a significant regulatory advantage.
The Artera algorithm was developed using data from thousands of patients and validated in multiple Phase 3 randomized controlled trials, including new data presented at the 2025 American Society of Clinical Oncology (ASCO) Annual Meeting in Chicago. Using the landmark STAMPEDE trial (NCT00268476), investigators found that the MMAI algorithm could identify which patients with high-risk, non-metastatic prostate cancer were most likely to benefit from adding abiraterone (Zytiga) plus prednisone to standard hormone therapy. Patients who were "biomarker positive" by the Artera algorithm and received abiraterone had a 5-year prostate-cancer-specific mortality of 9%, compared to 17% in similarly positive patients who received standard care. Among "biomarker negative" patients, abiraterone offered no detectable benefit. This kind of predictive capability — identifying who will and will not respond to a specific treatment — is the heart of precision medicine.
As of 2026, the ArteraAI Prostate Test has been incorporated into the NCCN Clinical Practice Guidelines for Prostate Cancer (V.4.2026), making it the first AI-based risk stratification tool recommended by NCCN for patients with localized prostate cancer. In June 2026, Artera also launched the ArteraAI Prostate Test for patients with metastatic hormone-sensitive prostate cancer (mHSPC), becoming the first externally developed prostate digital pathology algorithm deployed in the Tempus clinical genomics ecosystem. A registry study, the DIRECT-AI trial, is now enrolling patients specifically to document how the ArteraAI test influences real-world clinical decisions.
| Date | Company / Product | What It Does | Regulatory Path |
|---|---|---|---|
| Sept. 2021 | Paige AI — Paige Prostate | Cancer detection; flags suspicious foci on biopsy slides for pathologist review | De Novo (first-ever AI pathology product) |
| Feb. 2025 | Ibex Medical Analytics — Ibex Prostate Detect | Second-read safety net; catches missed cancers via AI heat maps | 510(k) clearance |
| Aug. 2025 | Artera — ArteraAI Prostate | Multimodal prognostication; predicts 10-year metastasis/mortality risk and therapy benefit | De Novo (new product category) |
The International Benchmark: The PANDA Challenge
Before any of these products reached the FDA, a landmark research competition established that AI could grade prostate cancer at pathologist-level accuracy across international borders. The PANDA (Prostate cANcer graDe Assessment) Challenge, published in Nature Medicine in January 2022, was the largest histopathology AI competition ever organized. A total of 1,290 developers from 65 countries submitted more than 17,000 algorithms trained on 10,616 digitized prostate biopsies. The winning algorithms were then validated on completely independent datasets from the United States and Europe.
On those external validation sets, the best algorithms achieved quadratic-weighted kappa values of 0.862 (U.S. cohort) and 0.868 (European cohort) in agreement with expert uropathologists — performance that matched or exceeded the inter-observer agreement between human specialists. Perhaps more importantly, PANDA found that AI algorithms outperformed general pathologists in grading reproducibility, even if specialist uropathologists still held a slight edge. The takeaway: in settings where specialist uropathology review is unavailable — a very common situation in community hospitals and in many parts of the world — AI can provide significantly more consistent grading than a non-specialist pathologist would.
The CONFIDENT-P trial, published in the Journal of Clinical Oncology Clinical Cancer Informatics in March 2025, took the next step: a prospective clinical implementation of the Paige Prostate AI in routine practice. The trial found that AI assistance improved pathologists' efficiency — shorter reading times, fewer requests for ancillary tests such as confirmatory immunostains — without any loss of diagnostic accuracy. This kind of real-world workflow validation, distinct from controlled research settings, is what moves AI from the laboratory to the pathology sign-out room.
The New Preprint: Grading Without Any Pathologist at All
Most of the AI tools described above are designed to assist a pathologist — flagging areas, reducing errors, improving speed. A new preprint study posted to medRxiv on June 24, 2026, from researchers at Brigham Young University and the company Pathtools.ai, pushes the question further: what happens if you remove the pathologist entirely?
The study applied a production AI system — PathTools Prostate v11.0, a gland-level diagnostic tool trained on needle core biopsy data — to 298 whole-slide images from 274 patients in the publicly available TCGA-PRAD (The Cancer Genome Atlas Prostate Adenocarcinoma) dataset. These were radical prostatectomy slides (from patients who had their entire prostate surgically removed), not biopsies — a more challenging domain shift for a system trained on biopsy tissue. The slides came from 25 different tissue source sites across the country, each with different digitization equipment and conditions. Critically, the AI system was applied completely "frozen" — without any tuning, calibration, or color normalization for the TCGA data. No pathologist annotated any region of any slide. The algorithm processed raw images from start to finish.
The study's primary question was not "does the AI agree with the pathologist?" but rather "can the AI grade alone, as the sole available predictor, stratify disease-free interval (DFI — a curated recurrence endpoint) as well as expert grading does?" This is the question that actually matters for the scenarios where AI would be most useful: remote or resource-limited settings where no pathologist exists to review the slide, or large archival research databases where systematic re-grading by expert pathologists is simply impractical.
The findings were striking. The automated grade group reproduced the pathologist's clinical grade at a quadratic-weighted kappa of 0.62 — right in the middle of the inter-observer range for specialist urological pathologists, and broadly consistent with published deep-learning graders validated against expert panels. The agreement was 48% exact (the AI and the pathologist gave the exact same grade group) and 86% within one grade group.
As a standalone predictor of disease-free survival, the automated grade group reached a Harrell's c-index of 0.69 (95% CI: 0.58–0.79). The pathologist's clinical grade group reached 0.78 (95% CI: 0.69–0.86). The difference was 0.09 — and when the researchers performed a rigorous paired bootstrap statistical test (comparing both methods on the same patients simultaneously), the confidence interval on that difference spanned zero. In statistical terms, the two methods were not separable at this sample size: one cannot conclude that the pathologist's grade was significantly more prognostically powerful than the AI's automated grade.
The continuous AI output — the percentage of tumor composed of high-grade Gleason 4 and 5 tissue — actually performed slightly better than the discretized grade group, reaching a c-index of 0.71 and proving robustly prognostic (hazard ratio 1.37 per standard deviation, p = 0.029). Adding pathologic T stage to the AI grade pushed the c-index to 0.74, approaching the performance of standard clinical nomograms like CAPRA-S — even though the TCGA dataset lacks PSA values and surgical margin status, which those nomograms also use.
— Ebbert, Szymansky, Perry, & Della Corte, medRxiv preprint, June 24, 2026
The study's authors are careful about what this finding does — and does not — mean. It does not prove that AI grading is superior to expert grading, or even exactly equivalent. The study had only 24 recurrence events, which limits statistical power to detect moderate differences. More importantly, the comparison was structurally unequal: the pathologist graded the entire prostatectomy specimen — all the tissue blocks, potentially including higher-grade tumor in other regions — while the AI graded only a single diagnostic slide per patient (the norm in the TCGA dataset). If the AI had access to all the same tissue blocks the pathologist reviewed, the gap might close further still.
The authors also identified two systematic errors in the AI's grading behavior. At the low end, the AI over-called low-grade disease: 23 of 30 clinical Grade Group 1 (Gleason 3+3) tumors were bumped up to Grade Group 2 by the AI, because even trace amounts of Gleason pattern 4 tissue — amplified by the domain shift between training data and prostatectomy slides — promoted the grade. At the high end, the AI under-called some GG5 tumors with cribriform-rich but pattern-5-poor morphology. The authors note that these errors largely collapse onto grade-group boundaries and do not substantially distort the underlying continuous high-grade fraction measurement, which is why reporting the continuous score (the percentage of high-grade tissue) rather than a discrete grade group is their recommended approach.
Important caveat: this is a preprint — it has not yet undergone formal peer review and should not be used to make clinical decisions. The study is also single-cohort, retrospective, with a relatively short median follow-up of 32 months.
What This Means for You as a Patient
For patients diagnosed today, these developments have several near-term implications worth understanding.
Your biopsy may already be read with AI assistance. If your biopsy was processed at a hospital or reference lab that has adopted Paige Prostate or Ibex Prostate Detect, an AI may have already flagged your slides — catching tiny cancer foci, confirming the presence of high-grade disease, or triggering a second look at an initially negative slide. This is happening now at leading academic centers and many commercial reference laboratories.
You may be offered an AI-based prognostic test. The ArteraAI Prostate Test is now both FDA-authorized and covered by Medicare (and many commercial health plans). It is recommended in NCCN guidelines for patients with localized prostate cancer. If you have intermediate-risk disease and are deciding between active surveillance and treatment, or weighing whether to add hormone therapy to radiation, asking your urologist or radiation oncologist whether an ArteraAI test is appropriate for your situation is a reasonable question. Results are typically available within one to two days of the lab receiving your biopsy specimen.
AI grading may be increasingly relevant for second opinions and archival research. If you received your original diagnosis at a small community hospital without specialist uropathology review, AI-assisted re-reading of your slides at a reference center is increasingly feasible. For patients enrolled in clinical trials or genomic classifier programs (like Decipher), the growing body of AI-graded archival data may ultimately allow better harmonization of risk stratification across studies.
AI does not yet replace the pathologist in clinical care. Every FDA-cleared product discussed here is explicitly designed as an assistive tool — it provides information to the pathologist, who retains final diagnostic responsibility. Regulatory and ethical frameworks in the United States and Europe do not currently permit purely autonomous AI diagnosis. The PathTools preprint is research demonstrating what is technically possible, not a product available for clinical use. The day of the fully pathologist-free diagnosis may come; it has not arrived yet.
What AI Does Not Yet Do Well
Honest accounting requires noting the limitations. Domain shift — performance degradation when AI systems are applied to slides from institutions, scanners, or tissue preparation protocols different from their training data — remains a real concern. The PathTools preprint demonstrated that this shift causes systematic grading errors even in a capable production system. Most leading commercial AI products require site-specific calibration (the AI learns the color profile and staining characteristics of the new laboratory's slides) before deployment, which can take weeks and requires a set of annotated reference cases from the new site. Systems that do not require this calibration — which PathTools demonstrated is at least possible — would be far easier to deploy in resource-limited settings, but they may trade accuracy for portability.
The TCGA data underlying the PathTools preprint also represents a particular challenge: prostatectomy tissue is more complex than needle core biopsies, and slides from 25 different institutional sources with heterogeneous digitization equipment represent a stress test that reflects real-world conditions. Not all AI systems will perform as well under these conditions.
Finally, no AI grading system has yet been shown to perform consistently and equally well across all racial and ethnic populations. Prostate cancer is significantly more prevalent and tends to be more aggressive in African American men — a difference with both genetic and socioeconomic dimensions. Most AI training datasets have historically skewed toward European ancestry cohorts. The Artera system explicitly reports consistent performance between African American and non-African American men in its validation data, which is a meaningful disclosure — but the field as a whole still has work to do on demographic equity in AI pathology validation.
Looking Ahead
The regulatory and research pipeline suggests the pace of development will only accelerate. As of mid-2026, the National Cancer Institute hosted a major workshop on digital pathology AI in clinical trials (published January 2026), highlighting both the promise of these tools for harmonizing grading across large multicenter studies and the remaining validation gaps. A new clinical trial — the PARADIGM trial, posted to ClinicalTrials.gov in June 2026 by University College London — is testing whether AI can match radiologists in interpreting prostate MRI scans, adding yet another dimension to fully AI-assisted prostate cancer diagnosis.
The BJU International published a comprehensive review in February 2026 ("Digital pathology-based artificial intelligence algorithms in prostate cancer: inside the 'black box'") examining how well current AI systems have been benchmarked against actual patient outcomes — metastasis and death — rather than just agreement with pathologists. That is precisely the type of rigorous outcome-based validation that the PathTools preprint begins to provide, and it is what regulators and clinicians will increasingly require before AI grades replace or supersede expert human grading in standard care.
For patients, the trajectory is reassuring: AI in prostate pathology is developing within a regulated framework, with peer-reviewed validation, clinical trials, and guideline endorsement setting the pace — not marketing hype. The goal is not to remove expertise from your care but to make expert-quality assessment available reliably, reproducibly, and everywhere — whether you are being biopsied at a major cancer center or a rural community hospital.
- Was AI used to assist in reading my biopsy slides? If so, which system?
- Was my biopsy reviewed by a specialist uropathologist, or by a general pathologist? If the latter, is a specialist second opinion appropriate?
- Am I a candidate for the ArteraAI Prostate Test? Is it covered by my insurance?
- How does my Gleason grade affect my eligibility for active surveillance?
- If I am enrolled in a genomic classifier program (like Decipher), has my Gleason grade been centrally re-reviewed?
- Ebbert JL, Szymansky J, Perry A, Della Corte D. "An Automated, Pathologist-free Gleason Grade Stratifies Disease-free Interval Comparably to Expert Grading from a Single Out-of-distribution Slide." medRxiv preprint. June 24, 2026. doi: https://doi.org/10.64898/2026.06.22.26356247
- Bulten W, Kartasalo K, Chen P-HC, et al. "Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge." Nature Medicine. 2022;28(1):154–163. doi:10.1038/s41591-021-01620-2. https://www.nature.com/articles/s41591-021-01620-2
- Paige. "Paige Receives First Ever FDA Approval for AI Product in Digital Pathology." Press release. BusinessWire. September 22, 2021. https://www.businesswire.com/news/home/20210922005369
- Ibex Medical Analytics. "Ibex Medical Analytics Receives First FDA 510(k) Clearance." Press release. BusinessWire. February 10, 2025. https://www.businesswire.com/news/home/20250210014120
- Pantanowitz L, Quiroga-Garza GM, Bien L, et al. "An artificial intelligence algorithm for prostate cancer diagnosis in whole slide images of core needle biopsies: a blinded clinical validation and deployment study." Lancet Digital Health. 2020;2(8):e407–e416. doi:10.1016/S2589-7500(20)30159-X
- Clarke H. "FDA grants 510(k) clearance to AI-powered prostate cancer diagnostic." Urology Times. February 3, 2026. https://www.urologytimes.com/view/fda-grants-510-k-clearance-to-ai-powered-prostate-cancer-diagnostic
- Artera. "Artera Receives U.S. FDA De Novo Marketing Authorization for AI-Digital Pathology Software Revolutionizing Prostate Cancer Care." Press release. BusinessWire. August 13, 2025. https://www.businesswire.com/news/home/20250813857439
- Saylor B, Clarke H. "FDA grants breakthrough device designation for ArteraAI Prostate tool." Urology Times. July 10, 2025. https://www.urologytimes.com/view/fda-grants-breakthrough-device-designation-for-arteraai-prostate-tool
- Clarke H. "FDA grants de novo authorization to ArteraAI Prostate." Urology Times. June 2026. https://www.urologytimes.com/view/fda-grants-de-novo-authorization-to-arteraai-prostate
- Artera. "The ArteraAI Prostate Test Enhanced with New Insights for Higher Risk Patients with Localized Prostate Cancer." Press release. BusinessWire. June 5, 2025. https://www.businesswire.com/news/home/20250605022847
- Parker C, Liu V, Mendes L, et al. "Multimodal artificial intelligence (MMAI) model to identify benefit from 2nd-generation androgen receptor pathway inhibitors (ARPI) in high-risk non-metastatic prostate cancer patients from STAMPEDE." Journal of Clinical Oncology. 2025;43(suppl 16). doi:10.1200/JCO.2025.43.16_suppl.5001
- Parker CTA, Mendes L, Liu VYT, et al. "External validation of a digital pathology-based multimodal artificial intelligence-derived prognostic model in patients with advanced prostate cancer starting long-term androgen deprivation therapy: a post-hoc ancillary biomarker study of four phase 3 randomised controlled trials of the STAMPEDE platform protocol." Lancet Digital Health. 2025. doi:10.1016/S2589-7500(25)00067-6. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(25)00067-6/fulltext
- Artera. NCCN Guidelines® for Prostate Cancer V.4.2026. Referenced at: https://artera.ai/for-clinicians
- Winayak R. "Study to evaluate impact of ArteraAI on clinical decision-making for prostate cancer [DIRECT-AI registry]." Urology Times. June 22, 2026. https://www.urologytimes.com/view/study-to-evaluate-impact-of-arteraai-on-clinical-decision-making-for-prostate-cancer
- Signify Research. "Digital Pathology News Round-up – M&A, AI Diagnostics & Market Trends (May 2026)." May 2026. https://www.signifyresearch.net/insights/digital-pathology-news-round-up-may-2026/
- Flach RN, van Dooijeweert C, Nguyen TQ, et al. "Prospective Clinical Implementation of Paige Prostate Detect Artificial Intelligence Assistance in the Detection of Prostate Cancer in Prostate Biopsies: CONFIDENT P Trial." JCO Clinical Cancer Informatics. 2025 Mar;9:e2400193. doi:10.1200/CCI.24.00193
- Chelebian E, Avenel C, Järemo H, et al. "Discovery of tumour indicating morphological changes in benign prostate biopsies through AI." Scientific Reports. 2025;15(1):30770. doi:10.1038/s41598-025-15105-6
- de la Calle CM, Baras AS, Lotan TL. "Digital pathology-based artificial intelligence algorithms in prostate cancer: inside the 'black box'." BJU International. 2026;137:596–604. doi:10.1111/bju.70164. https://bjui-journals.onlinelibrary.wiley.com/doi/10.1111/bju.70164
- Golatkar A, et al. [NCI Digital Pathology Workshop Report]. "Digital pathology imaging artificial intelligence in cancer research and clinical trials: An NCI workshop report." PMC. Published online January 2026. https://pmc.ncbi.nlm.nih.gov/articles/PMC12753273/
- PARADIGM Trial — "Prostate MRI Analysis by Radiologists and Artificial Intelligence — Disease Identification and Guided Management." ClinicalTrials.gov identifier: NCT07647445. Posted June 15, 2026. Lead Sponsor: University College London. https://clinicaltrials.gov/study/NCT07647445
- Allsbrook WC Jr, et al. "Interobserver reproducibility of Gleason grading of prostatic carcinoma: urologic pathologists." Human Pathology. 2001;32(1):74–80.
- Allsbrook WC Jr, et al. "Interobserver reproducibility of Gleason grading of prostatic carcinoma: general pathologist." Human Pathology. 2001;32(1):81–88.
- Imperial Prostate 6 — "Cancer Histology Artificial Intelligence Reliability Study." ClinicalTrials.gov identifier: NCT05228197. https://clinicaltrials.gov/study/NCT05228197
- Howard FM, et al. "The impact of site-specific digital histology signatures on deep learning model accuracy and bias." Nature Communications. 2021;12(1):4423. doi:10.1038/s41467-021-24698-1
- Epstein JI, et al. "The 2014 International Society of Urological Pathology (ISUP) Consensus Conference on Gleason Grading of Prostatic Carcinoma." American Journal of Surgical Pathology. 2016;40(2):244–252.
Comments
Post a Comment