SkinScreen

The SkinScreen application extends human capabilities in the detection and classification of skin lesions/skin cancer in the support value-based healthcare goals. SkinScreen offers the capability to detect malignant and benign skin lesions in real-time through a highly accurate and precise solution. The solution leverages the power of deep learning, a method under artificial intelligence (AI), to allow quicker and more accurate predictions than previously were available. Through a term that we have trademarked, called Indescribable Model, it is an AI model that is seeded with hyperparameters initially however the model continually trains itself in finding the best fit against the dataset without any future human intervention required. Currently, the detection is manually performed by a dermatologist or technician through a heuristic approach known as ABCDE (Asymmetry, Border Irregularity, Color, Diameter, Evolution).

SkinScreen offers a number of differences than other solutions in the market:

1. Ensure user privacy - By leveraging the latest MobileNetV2 architecture the AI model is able to run on a user's device and no images need to be uploaded back to SkinScreen's servers unlike other solutions.

2. Detect whether a skin lesion is present - Many AI skin detection solutions do not detect whether a skin lesion is present in the image initially. They rely upon manual intervention by the human user to provide a skin lesion image. For example, if a user provides an image of a giraffe their solutions will classify the image regardless. SkinScreen's sophisticated AI model is able detect whether a skin lesion is present prior to classification.

3. Detect more classes of skin lesions - By detecting 9 common benign and malignant classes of skin lesions (Actinic Keratoses, Angioma, Basal Cell Carcinoma, Dermatofibroma, Melanocytic nevus, Melanoma, Seborrheic keratoses, Squamous Cell Carcinoma, Vascular lesions)we are able provide better feedback for each individual who interfaces with SkinScreen. And we are continuing to expand the number of skin lesion classes that we support.

4. Provide higher accuracy and precision rates - We are leveraging a two-fold approach to accomplish the higher accuracy and precision rates. We first use a one-class classifier to identify whether a skin lesion is present in the image. If so, then we are able to provide back the 3 most likely skin lesion classes and their associated probabilities. Part of this is accomplished through the 180,000 images that we use to train our AI model.

5. Provide real-time feedback - SkinScreen is able to provide back results to the user in under two seconds on average. By leveraging MobileNetV2 architecture which has lower latency and higher accuracy and few proprietary enhancements we are able to notify the user of the results in a timely manner.

6. Provide user-friendly tools - SkinScreen's different platforms are able to assist users in their interactions with the tool. We try to accomplish this through support tools that are imperative in detecting skin lesions regardless of the user's background and skill sets.