Classification of Basal Cell Carcinoma using machine learning

Friday, February 12, 2016
Joshua Isaacson, Indiana Academy of Science, Indianapolis, IN
Over the past year I studied how to create and implement machine learning algorithms to classify images of random skin tumors as either a cancerous basal cell carcinoma (BCC) or a benign tumor. I created two computer programs to carry out this classification function, one centered around a logistic regression algorithm and the other utilizing a support vector machine (SVM). This investigation focused on obtaining the highest possible accuracy in classifying the random sets of images. After running the programs, I benchmarked the two programs’ accuracy with measures of specificity, sensitivity, and positive predictive value. I created and implemented the computer programs in Mathematica 10. My hypothesis: Implementing logistic regression and support vector machine algorithms will allow for a computer-based skin tumor classification system that has considerable accuracy. I concluded that in the near future, computer-based classification systems would reach a level of accuracy to a point where they could be a viable supplement to a physician’s diagnosis. The program setup around a support vector machine algorithm obtained a classification on the tumor sets of over 85%. These results are very encouraging and show that with further research, machine learning systems might possibly have a place in a medical environment soon. Further research will be done in fine tuning the support vector machine algorithm to better fit the classification problem at hand.