What's New In Digital Health

MetaOptima’s Research and Development team has been honoured with the ISIC 2018 Challenge for Skin Cancer Classification award. The team successfully earned first, second, and third place within the competition for their three submissions.

1. What is the ISIC Challenge?
The goal of ISIC 2018: Skin Lesion Analysis Towards Melanoma Detection is to help participants develop image analysis tools to enable the automated diagnosis of melanoma from dermoscopic images.

The ISIC 2018 challenge consisted of three tasks, with the MetaOptima Team focusing on the third titled “Task 3: Disease Classification”. The objective was to design a method  that can automatically diagnose an image of a skin lesion as one of the following seven skin diseases: melanoma, melanocytic nevus, basal cell carcinoma, actinic keratosis, benign keratosis, dermatofibroma, or a vascular lesion.

A set of 10015 images and diagnoses were provided for the training of algorithms, however participants had the option to use any other external data.  For the final evaluation, participants made predictions on a held-out set of 1512 images and submitted them to an online system to be scored. The metric used for scoring was the accuracy of the classifier on each of the classes averaged together, known as “mean recall” or “balanced accuracy”.

2. MetaOptima’s achievements
MetaOptima’s submissions ranked first, second, and third for its approaches titled Top 10 Models Averaged (88.5%), Meta Ensemble (88.2%), and Best Single Model (87.1%) respectively. Primary Metric Values indicate that the company’s entries ranked notably higher when compared to submissions with a difference of 2.9% between MetaOptima’s top ranking approach and the fourth placed entry.

Although this model does not constitute a real-world use case as presented, the approach offers substantial clinical value as it proved the possibility to obtain high classification accuracy by ensembling multiple models trained for a 7-class skin lesion classification. The outcome of this competition was especially encouraging because it demonstrates how our software’s AI-based algorithms possess significant potential to support medical professionals with evidence-based clinical decisions for enhanced patient care. Challenges such as these truly highlight the future for artificial intelligence in dermatology in how can be utilized to empower doctors and care providers to supply patients with improved  treatment options.