Train AI model to assist doctors to analyze images of multiple skin diseases captured from portable device as decision support. 
The training and validation data is based on pathology report as ground truth . After verification, doctors label images on pathological features. 
The AI model algorithm reaches over 90% in F1-score.

The AI training and validation data is based on pathology report as ground truth.
It adopts Inception V3, Inception ResNet V2, and NASNet three models.
After the models reach over 80% in F1-score, it applies Ensemble Voting techniques to enhance the performance and it reaches over 90% F1-score.

Many dermatological diseases can hardly be diagnosed by naked eyes and therefore pathological examination will be frequently needed.
To goal to train AI is to ease the doctor＇s workload and enhance diagnosis quality by reducing the frequency rate of invasive examination.