Measurement of psoriasis affected area with an artificial neural network: Supplementary Material

Published: 30 August 2022| Version 4 | DOI: 10.17632/8nzj9y2rhm.4
Contributors:
Woo Hyup Lee,
Sungmin Lee,
Jiho Kim,
Ju Hee Han,
Yeong Ho Kim,
Jun Hwi Kim,
Ji Hyun Lee,
Young Min Park,
Chul Hwan Bang

Description

Supplementary Table 1 shows performance comparison with existing methods. Regardless of methodology used, APD showed the best performance. The detailed method is presented with a table. Supplementary Figure 1 shows the flow chart of dataset configuration. The data were obtained from 149 patients who visited Seoul St. Mary’s Hospital Dermatology Clinic from 2018 to 2020, and mainly consisted of images of each area division used to calculate PASI score (i.e., trunk, upper extremities, lower extremities). Various plaque psoriasis phenotypes (small plaque, large plaque, regressed lesion), non-psoriatic lesion, and artifacts (e.g., accessories, shadows) were included in the dataset. The 35 excluded images were those of a face or out of focus. To evenly distribute the patient severities in the training and test sets, the data set was first split into three groups (mild, moderate, severe) by PGA. Random sampling was performed in each group in a 4:1 ratio. To evaluate the PASI area score, inappropriate images for use with the palm method were excluded, resulting in 297 images to determine the performance of the three dermatologists. With respect to input image, APD determines skin area and psoriatic lesion to calculate the percentage of affected area. “Ground truth” was obtained by manually annotating each lesion margin with agreement between three annotator dermatologists. “Skin inference” was the result of detected skin area by the automated psoriasis detector, and “prediction” was the detection of a psoriatic lesion by the automated psoriasis detector. Supplementary Figure 2 shows the flowchart of the APD. The Cascade Mask R-CNN with a Swin Transformer small (Swin-S) backbone, the same model used in APD, was used to determine affected skin area with pretrained weights from the COCO dataset. To calculate skin area precisely, the patient needs to be fully unclothed. Few cases were insufficiently unclothed, leading to imprecise PASI area score and inaccurate percentages of psoriasis-affected area. In such cases, an additional skin detection method (skin color-based thresholding) was applied. Supplementary Figure 3 shows an example of automated psoriasis detector implementation. Using the input image, APD determines skin area and psoriatic lesion to calculate the percentage of affected area. “APD inference” is the process of determining the presence of a psoriatic lesion by APD. “Skin area inference” is the process of detecting skin area by APD. “Skin area cleansing” involves application of an additional skin detection method (skin color-based thresholding) due to artifacts. “Find the lesion ratio” is the calculation of the percentage of psoriasis affected area versus skin area. Ground truth was obtained by manually annotating each lesion margin with agreement between three annotator dermatologists. APD, Automated psoriasis detector.

Files

Institutions

Catholic University of Korea, Seoul Saint Mary's Hospital

Categories

Artificial Intelligence, Psoriasis

License