Skin--Cancer--Diagnosis

Skin cancer is a prevalent cancer that significantly contributes to global mortality rates. Early detection is crucial for a high survival rate. Dermatologists primarily rely on visual inspection to diagnose skin cancers, but this method is inaccurate. Deep learning algorithms can enhance the diagnostic accuracy of skin cancers. However, these algorithms require substantial labeled data for effective training. Acquiring annotated data for skin cancer classification is time-consuming, expensive, and necessitates expert annotation. Moreover, skin cancer datasets often suffer from imbalanced data distribution.
Generative Adversarial Networks (GANs) can be used to overcome the challenges of data scarcity and lack of labels by automatically generating skin cancer images. However, training and testing data from different distributions can introduce domain shift and bias, impacting the model’s performance. This dissertation addresses this issue by developing deep learning-based domain adaptation models.
Additionally, this research emphasizes deploying deep learning models on hardware to enable real-time skin cancer detection, facilitating accurate diagnoses by dermatologists. Deploying conventional deep learning algorithms on hardware is not preferred due to the problem of high resource consumption. Therefore, this dissertation presents spiking neural network-based (SNN) models designed specifically for hardware implementation. SNNs are preferred for their power-efficient behavior and suitability for hardware deployment.

Skin cancer is a major medical problem. If not detected early enough, skin cancer like
melanoma can turn fatal. As a result, early detection of skin cancer, like other types of
cancer, is key for survival. In recent times, deep learning methods have been explored to
create improved skin lesion diagnosis tools. In some cases, the accuracy of these methods
has reached dermatologist level of accuracy. For this thesis, a full-fledged cloud-based
diagnosis system powered by convolutional neural networks (CNNs) with near
dermatologist level accuracy has been designed and implemented in part to increase early
detection of skin cancer. A large range of client devices can connect to the system to
upload digital lesion images and request diagnosis results from the diagnosis pipeline.
The diagnosis is handled by a two-stage CNN pipeline hosted on a server where a
preliminary CNN performs quality check on user requests, and a diagnosis CNN that
outputs lesion predictions.