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

E-Poster Display

1101P - Near real-time intraoperative melanoma diagnosis using deep neural networks

Date

17 Sep 2020

Session

E-Poster Display

Topics

Tumour Site

Melanoma

Presenters

Ruple Jairath

Citation

Annals of Oncology (2020) 31 (suppl_4): S672-S710. 10.1016/annonc/annonc280

Authors

R. Jairath1, N.K. Jairath1, T. Tejasvi2, L.C. Tsoi2

Author affiliations

  • 1 Oncology, Rogel Cancer Center, 48019 - Ann Arbor/US
  • 2 Department Of Dermatology, University of Michigan, 48109 - Ann Arbor/US

Resources

Login to get immediate access to this content.

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

Abstract 1101P

Background

Intraoperative diagnosis is essential for providing safe and effective care during staged melanoma resections. The existing workflow for intraoperative diagnosis based on H&E staining of processed tissue is time-, resource-, and labor-intensive. Moreover, interpretation of intraoperative histologic images is dependent on a pathology workforce that is contracting and unevenly distributed across the centers where melanoma excisions are performed worldwide.

Methods

We developed an automated workflow that uses deep convolutional neural networks (CNN) to predict diagnosis at the bedside in near real time. Specifically, our CNN, trained retrospectively on over 10,000 H&E images that were collected prospectively across multiple institutions, predicts brain tumor diagnosis in the operating room in under 150 s, which is an order of magnitude faster than conventional techniques.

Results

Our trained CNN was compared to pathologist-confirmed diagnoses in a testing set of 204 patients. Primary endpoint was overall diagnostic accuracy. We show that CNN-based diagnosis of H&E images was noninferior to pathologist-based interpretation of histologic images (overall accuracy, 94.6% vs 95.5%). Additionally, our CNN learned a hierarchy of interpretable histologic feature representations to classify the major histopathologic classes of melanoma. We then developed and implemented a semantic segmentation method that can identify tumor infiltrated and diagnostic regions within the images. Mean intersection over union values was 61 ± 28.6 for ground truth diagnostic class and 86.0 ± 28.6 for tumor-infiltrated regions.

Conclusions

We have demonstrated how the application of deep learning can result in near real-time intraoperative melanoma diagnosis. Our workflow provides a means of delivering expert-level intraoperative diagnosis where dermatopathology resources are scarce and improve diagnostic accuracy and efficiency in resource-rich centers.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

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.