Creating a Haptic 4D Model Along With Machine Learning Analysis by Developing a Non-Invasive Pressure Mapping Method to Screen Genital Skin Cancer

Abstract

Cancer is Curable, But only when it is detected in its early stages. When focusing on genital skin
cancer matters become worse since factors like privacy, comforts, social hurdles and restrictions play a key role
in postponing cancer detection in genital regions. Biopsy is an invasive method to accurately screen cancer, but
when conducted in genital region causes pain, infection, numbness etc. Hence, our engineering goal is to screen
suspicious skin lesions non invasively providing detailed analysis to doctors virtually, reducing the number of
times patients experience an invasion of their privacy, giving them control of the screening process, respecting
their privacy and promoting early detection.
We developed a machine learning model, executed and deployed as a mobile app. The image of the lesion is
processed and fed into our Deep Convolutional Neural Network (DCNN) - trained and tested with 5000 images
- yielding a percentage probability report of the lesion being classified as Malignant, benign or premalignant. If
malignant, further classifying them within the 5 main skin cancers (Aktinik Keratosis, Squamous Cell
Carcinoma, Melanoma,, Basal Cell Carcinoma and Intraepidermal Carcinoma) with an accuracy of 83% on
confusion matrix.
For further analysis, the patient uses a pressure mapping kit and applies it on the lesion, which delocalizes a
Non-Newtonian Fluid. Using gradient localization methods we map gel density against pressure to create a 3D
support’s file which is convoluted with the 3D STL file generated by the lesion’s top image to produce a single
3D flexible printing file, when printed by the doctor gives a flexible haptic model which provides accurate
tactile feedback.