More than 600 hours of processing time invested in training and improving neural networks.
Identify different types of benign & malignant spots in the skin, just by taking a photo or uploading an image (ATTENTION: THIS IS NOT THE RECOMMENDED WAY TO IDENTIFY A TYPE OF BLEMISH, CONSULT A CERTIFIED DERMATOLOGIST FOR A RELIABLE DIAGNOSIS).
If you decide to use this app, it is highly recommended to load a photo made with a dermatoscope as regular pictures don't show small details (THIS APP IS NOT FOR USING "YOUR SMARTPHONE AS A DERMATOSCOPE", THE DERMATOSCOPE IS AN EXTERNAL PROFFESSIONAL TOOL YOU COULD USE TO OBTAIN HIGH QUALITY IMAGES).
About how this app is made and technical specifications of the Convolutional Neural Networks involved (for Software Scientists interested in this field):
This application do not use only rotated large scale images to make the identification. In this app well-designed data augmentation techniques are used, WICH NOT GUARANTEES THE APP TO BE PRECISE IN THE RESULTS, CONSULT A DERMATOLOGIST IF YOU HAVE ANY DOUBT.
Resulting validation general accuracy: 70.5% (note that a random result will obtain 12.5% accuracy; in a basic Melanoma-Not Melanoma model it's 50.0%, this is not the case, the model was also trained with 5 instead of 1 different detail levels, reducing the bias but resulting in less validation accuracy).
As this is a skin cancer identification App that uses the usual repositories used in this type of applications, THIS APP SHOULD BE FREE, as it is not allowed to monetize the use of this data.
Identify different types of benign & malignant spots in the skin, just by taking a photo or uploading an image (ATTENTION: THIS IS NOT THE RECOMMENDED WAY TO IDENTIFY A TYPE OF BLEMISH, CONSULT A CERTIFIED DERMATOLOGIST FOR A RELIABLE DIAGNOSIS).
If you decide to use this app, it is highly recommended to load a photo made with a dermatoscope as regular pictures don't show small details (THIS APP IS NOT FOR USING "YOUR SMARTPHONE AS A DERMATOSCOPE", THE DERMATOSCOPE IS AN EXTERNAL PROFFESSIONAL TOOL YOU COULD USE TO OBTAIN HIGH QUALITY IMAGES).
About how this app is made and technical specifications of the Convolutional Neural Networks involved (for Software Scientists interested in this field):
This application do not use only rotated large scale images to make the identification. In this app well-designed data augmentation techniques are used, WICH NOT GUARANTEES THE APP TO BE PRECISE IN THE RESULTS, CONSULT A DERMATOLOGIST IF YOU HAVE ANY DOUBT.
Resulting validation general accuracy: 70.5% (note that a random result will obtain 12.5% accuracy; in a basic Melanoma-Not Melanoma model it's 50.0%, this is not the case, the model was also trained with 5 instead of 1 different detail levels, reducing the bias but resulting in less validation accuracy).
As this is a skin cancer identification App that uses the usual repositories used in this type of applications, THIS APP SHOULD BE FREE, as it is not allowed to monetize the use of this data.
Show More
