Application of U-Net with Inceptionv4 Encoder for Localizing and Counting Monkeypox Lesions in Patient Photographs![Download PDF](/images/pdf_icon_blue.svg)
Keywords: Monkeypox, segmentation, deep learning, U-Net, lesion counting
TL;DR: First study of monkeypox lesion counting from patient photographs shows promising results using a deep U-Net.
Abstract: Monkeypox is a disease caused by infection with monkeypox virus that causes significant morbidity in several central and western African countries. There are currently no proven, safe treatment for monkeypox virus infection. Lesions pass through several stages before resolution. Evaluating the clinical efficacy of possible treatments involves manually tracking changes in lesion counts over time until lesion resolution (scabbed or desquamated), which is both labor intensive and prone to human error. To support a randomized controlled trial evaluating putative therapeutics we developed a deep learning method for monkeypox lesion localization and counting in patient photos. In 20 photos from 12 patients with monkeypox, we manually annotated all visible lesions and trained a U-Net network with Inceptionv4 encoder to localize and count the lesions. In a leave-one-out evaluation our method shows promising results for lesion segmentation, with a median Dice of 0.74 (interquartile range: 0.72, 0.79) on the unseen photos. Automated lesion counting was evaluated on a second held-out set of 20 photographs. Automatic lesion counting performs similarly to human raters when compared to ground truth, with median lesion count difference of 2.00 (-8.00, 10.25) for the algorithm compared to 5.50 (2.75, 12.00) and -2.00 (-7.00, 1.25) for two other human raters. The variability by mean pairwise difference U-statistic of lesion count between the algorithm and by the ground truth was 24 lesions.
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: validation/application paper
Primary Subject Area: Segmentation
Secondary Subject Area: Application: Other
Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.
0 Replies
Loading