Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

ePoster Display

1144P - When AI-based image analysis gets in clinical trials

Date

16 Sep 2021

Session

ePoster Display

Topics

Pathology/Molecular Biology

Tumour Site

Breast Cancer

Presenters

Paul Mésange

Citation

Annals of Oncology (2021) 32 (suppl_5): S921-S930. 10.1016/annonc/annonc707

Authors

P. Mésange1, M. Bonnet2, J. Deen3

Author affiliations

  • 1 Aph Department, Covance Central Laboratory Services SA, 1217 - Meyrin/CH
  • 2 Histology, Covance Central Laboratory Services SA, 1217 - Meyrin/CH
  • 3 Histology, Covance, Geneva/CH

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 1144P

Background

Especially during the COVID-19 pandemic, implementation and development of whole slide scanning lead to an active growth of digital pathology and image analysis (IA). In the present study, we retrospectively conducted a global evaluation of three Breast cancer markers: ER, PR and Ki67 with the aim to study the correlation between pathologist conventional semi quantitative scoring method on glass and scanned slides versus artificial intelligence-based IA.

Methods

Study samples were scored independently either by five independent pathologists on scanned images and glass slides, or using supervised IA algorithms (IA results validated by pathologists). The readout for the three markers was the percentage of tumor positives stained cells. The correlation between the pathologist evaluation on glass slides versus scanned images was calculated using Pearson’s correlation coefficient. Pathologist’s evaluations and IA results were compared using Intraclass Correlation Coefficient (ICC). Additionally, the average time spent by the pathologist per sample was measured for each evaluation method.

Results

The correlation of pathologist evaluation between glass slide and scanned image showed a Pearson’s correlation coefficient ≥ 0.90 for each marker. The ICC between IA algorithm and pathologist was on average over 0.8 for the three markers, showing a good agreement between the different scoring method. However, some challenges were identified related to the detection of tumor area that needed some additional pathologist review for specific complex cases. Overall, time required by the pathologist for a complete evaluation decreased by roughly 3 times when supported by image analysis tools.

Conclusions

Based on the Pearson’s correlation coefficient and the ICC results, we observe an equivalence in the pathologist conventional scoring (Image or glass slides) and the use of IA. In an era where regulations are still being discussed for the use of algorithm by the FDA (AI-Based or not), we can mitigate regulatory requirements by having pathologists reviewing the results of a digital analysis. We conclude here to a benefit from the combination of pathologist evaluation and IA in terms of time with at least equivalent results in terms of accuracy.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Covance.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.