Suspicious Moles Detection Using Deep Learning

Melanoma skin cancer is common, deadly and diagnosis today requires manual examination by a dermatologist. For any suspicious mole on the patient’s body, the dermatologist should look at through a dermoscope (polarized light photography) and this is the only way to determine later whether or not that mole is malignant. Marpe Technologies develops a system that scans the patient’s body in visible light using a high-resolution camera, finds suspicious moles and sends them to be inspected using a dermoscope. This is supposed to decrease 'doctor time'. The aim of the project was to classify "suspicious moles" through visible light photography, the classification should be automatic, at technician level and beat 93% accuracy.
We chose to address the problem with CNN and train it on patches of suspicious and not suspicious body areas. A suspicious area is an area that contains a suspicious mole that should be inspected by a dermatologist with a dermoscope. The final accuracy of the network is 99.8% and its precision is 92.7%.

mole inspection
Suspicious Moles Detection Using Deep Learning
Suspicious Moles Detection Using Deep Learning
Collaboration: