Skin Cancer Screening Network (SkinCancerScreenNet) is a deep convolutional neural network that accepts skin lesion dermoscopy images of type jpg as the input, and predicts the likelihood of the lesion being benign or malignant as the output. This particular model can be used on skin dermoscopy images, and can be used on a single image, or sets of images.
DarwinAI does not endorse any specific procedures, products, tests or treatments that may or may not be related to the use of the Model.
If you or someone you know is having a medical emergency, please immediately contact the authorities, a doctor, or a hospital.
The information provided here is not considered Protected Health Information as defined by HIPAA and HIPAA does not apply to the results or any collection of information.
The indicated performance of the Model is based on various tests run by DarwinAI and performance is not guaranteed.
Create a Modzy account to get started →
The model was validated on an 442 image subset of the ISIC dataset, with 221 images of malignant melanoma, and 221 images of benign nevi. On this test dataset, SkinCancerScreenNet achieved a sensitivity of 92%, and a specificity of 77%.
79.9% Positive-Predictive-Value (PPV) – A higher precision score indicates that the majority of labels predicted by the model for different classes are accurate.
91.9% Sensitivity – A higher recall score indicates that the model finds and predicts correct labels for the majority of the classes it is supposed to find.
SkinCancerScreenNet was developed using a human-machine collaborative design strategy, where human-driven principled network design prototyping is combined with machine-driven design exploration to produce a network architecture tailored specifically for skin cancer detection. SkinCancerScreenNet is comprised of just 520K parameters while still achieving high performance levels, and was tuned for a higher sensitivity level.
The model is implemented in Tensorflow.
The model was trained using the ISIC dataset, an open access dataset consisting of over 23,000 skin lesion images. The model was trained on a 21,659 image subset of the ISIC dataset.
Note that it is recommended to use a GPU when performing inference with this model.
See how quickly you can deploy and run models, connect to pipelines, autoscale resources, and integrate into workflows with Modzy—the ModelOps and MLOps platform