This summer, I've created an automated program using JSL scripting and JMP modeling that analyzes a given skin lesion to determine whether it could be malignant (skin cancer). My program uses image segmentation techniques to extract numbers representing the main variables of asymmetry, border irregularity, color variation, and texture from the image, and uses discriminant analysis to compare the given lesion's metrics to those of 153 lesions in the accompanying data set to make a conclusion. With these samples, the current accuracy is 95%, which means there is room for error - note that the add-in's diagnosis does not in any way replace an official diagnosis from a dermatologist or doctor. I've had a blast creating this project, and I've learned a great deal from it. Please feel free to comment or email me thoughts you have about my idea - I'd love to hear your suggestions and feedback.
I will also be presenting a poster on this project at JMP Discovery Summit in September, so please stop by if you're planning to attend!
I'd especially like to thank John Ponte and for their mentorship and guidance, and Diana Levey and the JMP marketing team for their support and encouragement.
