Final Results
Testing Data
As far as results went, the detector was fairly accurate with our testing images. We tested a total of 20 different images, 14 melanoma and 6 moles, and were 80% (16/20) accurate during those tests. The chart of what images were correct on and missed is shown to the right. The dataset that contains our testing images (and their citations) is linked below.
Image Courtesy of: Ryan Downey
Image Courtesy of: Ryan Downey
Potential Segmentation Issues
Issues that were stated in the Segmentation portion our website could have caused some false positives within our detection results.
Sample Image (Melanoma11.jpg)
1. Asymmetry Index = 19.0278
2. Circulatory = 1.1545
3. Color Variation = 22.1014
​
Asymmetry Index and Color Variation are both greater than their thresholds so the algorithm classifies the image as melanoma.
Image Courtesy of: https://www.health.com/condition/melanoma/is-it-a-mole-or-skin-cancer