Error rates in reporting prostatic core biopsies

Oxley, Jon; Sen, Chandan K.

doi:10.1111/j.1365-2559.2011.03794.x

Cited by 14 publications

(22 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our results, we show that the unreliable (empty or multiple) predictions for prostate cancer detection were 22% at a confidence level of 99.9%. This represents an error rate of 0.1%, which is markedly lower than the 2% error rate that has been reported for pathologists 13,14 . Our results on Test set 6, containing unusual morphological patterns, demonstrate how CP can identify cases that the underlying AI system had little or no exposure to during training.…”

Section: Discussioncontrasting

confidence: 56%

See 1 more Smart Citation

Estimating diagnostic uncertainty in artificial intelligence assisted pathology using conformal prediction

Olsson

Kartasalo²,

Mulliqi³

et al. 2022

Nat Commun

View full text Add to dashboard Cite

Unreliable predictions can occur when an artificial intelligence (AI) system is presented with data it has not been exposed to during training. We demonstrate the use of conformal prediction to detect unreliable predictions, using histopathological diagnosis and grading of prostate biopsies as example. We digitized 7788 prostate biopsies from 1192 men in the STHLM3 diagnostic study, used for training, and 3059 biopsies from 676 men used for testing. With conformal prediction, 1 in 794 (0.1%) predictions is incorrect for cancer diagnosis (compared to 14 errors [2%] without conformal prediction) while 175 (22%) of the predictions are flagged as unreliable when the AI-system is presented with new data from the same lab and scanner that it was trained on. Conformal prediction could with small samples (N = 49 for external scanner, N = 10 for external lab and scanner, and N = 12 for external lab, scanner and pathology assessment) detect systematic differences in external data leading to worse predictive performance. The AI-system with conformal prediction commits 3 (2%) errors for cancer detection in cases of atypical prostate tissue compared to 44 (25%) without conformal prediction, while the system flags 143 (80%) unreliable predictions. We conclude that conformal prediction can increase patient safety of AI-systems.

show abstract

Section: Discussioncontrasting

confidence: 56%

“…Errors in prostate cancer diagnosis are rare but well recognized 13,14 . It is clear that there is less acceptance of machine learning mistakes than of human mistakes 15 .…”

Section: Discussionmentioning

confidence: 99%

Estimating diagnostic uncertainty in artificial intelligence assisted pathology using conformal prediction

Olsson

Kartasalo²,

Mulliqi³

et al. 2022

Nat Commun

View full text Add to dashboard Cite

show abstract

“…While very few studies are devoid of selection bias and quoted rates vary from 0% to 10%, most investigators report an average failure rate of 2% to 3%. [8][9][10][11][12] We postulated that prospective multiplex IHC could improve cancer pick-up rate and our audit data shows that, following its introduction, our cancer positivity rate increased from 44.1% in 2004-2009 to 51.7% in 2010-2013. 7 The combination of prospective multiplex IHC and TPMbx, including magnetic resonance imaging before biopsy, further improved the cancer detection rates.…”

Section: Discussion Prospective Multiplex Ihc Improves Cancer Detectionmentioning

confidence: 88%

Use of Prospective Multiplex Immunohistochemistry to Redefine Tissue Pathways of Diagnostic Core Biopsy of Prostate

Thomas

Wedden

Afzal

et al. 2022

Applied Immunohistochemistry &Amp; Molecular Morphology

View full text Add to dashboard Cite

The incidence of prostatic cancer in the United Kingdom has increased over 40% in the past 30 years. The majority of these cancers are diagnosed by core biopsy, posing a considerable strain on a service that struggles to recruit sufficient histopathologists. The current methodology for tissue diagnosis has a significant false-negative rate, small false-positive rate, and a proportion of indeterminate diagnoses. Therefore, this area presents an opportunity both to improve diagnostic quality and to reduce the burden on resources. We investigated streamlining tissue pathways by increasing the utilization of readily available resources to reduce the burden on scarce resources and improve the accuracy of diagnosis. This involved applying prospective multiplex immunohistochemistry (IHC) using 4 different markers (CK5, p63, racemase, and Ki-67) and 2 chromogens. We conducted a prospective study using over 8000 cores and 3 consultant histopathologists. The pathologists assessed each core using either conventional stains (hematoxylin and eosin) only or multiplex IHC only. The results of this assessment were later compared with the overall assessment made for the final histologic diagnosis. Results show that IHC alone has a positive predictive value of 98.97% and a negative predictive value of 99.91%, while hematoxylin and eosin alone has a positive predictive value of 94.21% and negative predictive value of 99.07%, demonstrating improved diagnostic accuracy. When assessed against the use of on-demand IHC, prospective IHC improves turn-around-times, reduces indeterminate diagnoses, improves pathologist’s accuracy and efficiency and, in overall terms, is cost-effective. In addition, it is possible to structure these tests within the routine of a diagnostic service with little impact on the overall capacity of the laboratory.

show abstract

“…Oxley et al. showed that non‐specialists made more errors than the urological pathologists in reporting prostatic core biopsies. Lee et al.…”

Section: Discussionmentioning

confidence: 99%

Can the Leibovich score for clear cell renal cell carcinoma (ccRCC) be accurately reported by a general pathologist?

et al. 2013

View full text Add to dashboard Cite

Objectives• To evaluate the accuracy of reporting of the Leibovich score (LS) by general pathologists at our centre by comparing the LS in their initial reports with the LS in subsequent review reports by our specialist urological pathologists.• We also assessed whether a revision in the LS subsequently altered the risk group. Patients and Methods• In all, 54 consecutive patients had had their initial histology after nephrectomy reported by a general pathologist between August 2008 and March 2012.• The histological slides were subsequently re-reviewed by the two specialist urological pathologists and revised LS were given if appropriate. The cases were then discussed at the Uro-oncology Multi disciplinary meeting (MDT) and the reason for the revised LS was discussed with the participating panel and reviewed by the uropathologist. Results• The general pathologists allocated the 54 patients into low-(13 patients), intermediate- (25) and high-risk (16) categories.• After a review of the slides by the specialist urological pathologists and discussion at the MDT meeting the LS was revised in 23 patients (42.6%).• The 23 patients who had their LS revised were divided into two groups for the purpose of the present study. On revision of the LS a total of eight patients (14.8%) changed their prognostic group. • The change in the prognostic group in the eight patients was increased from intermediate-to high-risk in four patients.• There was a reduction in the LS in four patients, which was from high-to intermediate-risk category in one patient and intermediate-to low-risk in three patients. Conclusion• In the present study, histopathology review by a specialist urological pathologist led to a change in LS in 42.6% of cases leading a change in risk grouping in 14.8% of cases. This level of discrepancy is not insignificant. However, the few cases evaluated in the present study does limit robust conclusions and further studies are needed to investigate this issue, so that recommendations can be made to enhance diagnostic accuracy and reproducibility.

show abstract

Error rates in reporting prostatic core biopsies

Cited by 14 publications

References 17 publications

Estimating diagnostic uncertainty in artificial intelligence assisted pathology using conformal prediction

Estimating diagnostic uncertainty in artificial intelligence assisted pathology using conformal prediction

Use of Prospective Multiplex Immunohistochemistry to Redefine Tissue Pathways of Diagnostic Core Biopsy of Prostate

Can the Leibovich score for clear cell renal cell carcinoma (ccRCC) be accurately reported by a general pathologist?

Contact Info

Product

Resources

About