Dermoscopy AI Demo Guide

1. Go to https://demo.belle.ai and select the AI module.

2. Upload a corresponding image. You will have the option to add multiple images (look for the plus sign that will appear after the first image is uploaded).

3. The Dermoscopy AI will then analyze the image, providing an estimated risk level.

• The risk level is calculated based on the scores of each disease class and subclass going from 0 to 5 and highlights the probability of skin cancer in the image:
  • 0: " Cancer Signs not Found"
  • 1: " Probably Benign but Monitor"
  • 2: “Check More to Rule Out Cancer”
  • 3: " Pre-Cancer or Uncertain"
  • 4: " Risk of Cancer"
  • 5: " Risk of High-Grade Cancer"
• Click on the ‘ABCD’ button or ‘i’ button in the top right corner to see the ABCD explanation or risk level key, respectively.
• The numbers under “Comparative Images” are Belle Index Matching (BIM) scores. Such scores indicate how much a given image compares to other images of the displayed disease classes or subclasses. The higher the BIM score, the stronger the correlation between the given image and the database images for a particular condition.
• You may add notes or comments under “What do you think about the result?”
• You may check the box of what condition you believe is present. Your input will not change any scoring.

• Under Shape, the green lines are the symmetric axes of shape/ color if any. In case of Asymmetry, there is no line.
• Under Borders, the green plus signs indicate the concave points of the lesion.

• Under Color, the right pixelated image is the raw result of image computing to find the closest matching color. The original image is divided into chunks where the color is evaluated against a list of 9 potential colors: black, blue, red, light brown, dark brown, white, orange, purple, and yellow. This pixelated image is then used to draw colored contours on top of the original image which is the left image.
• Under Diameter, the image scale, mole area, and mole size are given.

• Under Network Binary, there is a computed image of the lesion that extracts the pigmented patterns inside the lesion.
• Under Inverted Network Binary, there is a computed image of the lesion that extracts the non-pigmented patterns inside the lesion.

4. You may revisit previous analyses via the History tab (located on the sidebar) which automatically saves previous results (yet comments on such analyses must be saved manually by clicking the save button).

Questions or Feedback?