AI Daily - Artificial Intelligence News

View Original

AI That Can Detect Skin Cancer

Melanoma skin cancer is one of the most common types of cancer in the UK. According to Cancer Research UK, there are around 16,200 cases each year and since the early 1990s, skin cancer incidence rates have more than doubled. Before the situation becomes increasingly difficult to handle, what is AI doing to counter it?

Melanoma is a type of skin cancer that can develop from the sun's UV radiation, which makes skin cells grow abnormally. Melanoma is curable and if detected early, there is a 97% chance of survival. Yet finding melanoma can be difficult, as signs may not be easy to notice. A harmless-looking spot on one's leg may be a benign skin lesion or growing cancer cells. But generally, the spread of melanoma takes the appearance of a mole changing shape or even becoming bigger.

Skin Analytics is a research-led firm founded in 2012, and its focus is to detect melanoma skin cancer as soon as possible. The firm provides its services in the form of an app and uses a deep learning system called DERM (Deep Ensemble for the Recognition of Malignancy) AI which is based on convolutional neural networks. By simply taking a photo, the customer can keep track of their skin history over time. And if there are suspicious changes to their skin, the customer will then be able to share those details with their local healthcare provider.

Convolutional Neural Networks, also known as CNNs, are a class of neural networks that run by analysing images. A report in 2017 was released by Stanford University, showing students building a skin cancer classification tool that used CNNs, just like Skin Analytics' app. For the many applications of AI, the base model, for whatever program, must be trained. CNNs are no different. The students used 129,450 images, which consisted of 2032 different diseases to train the deep learning model - so it would be able to distinguish between the types of cancer cells. The model was able to compete on par with professional dermatologists and even performed better in certain tests, which demonstrated its capability to detect skin cancer.

Even though AI in dermatology seems to be progressing, some challenges are faced in the clinic. For the AI to analyse the picture it is given, the picture must be of very high quality. Hospitals may not have the facilities to provide such high-quality images that are all uniform, for the model to recognise. Another issue, in general, is that the AI used in dermatology isn't capable enough to diagnose a spread of diseases. It only detects more common ones. This may lead to an inaccurate diagnosis.

Though there have been many developments in the field of dermatology regarding AI, there are still pressing issues that need to be resolved. For now, it is better to leave the job to a dermatologist as many medical factors (weight, skin colour etc..) are involved before giving a proper and accurate diagnosis. AI isn't at the stage yet where it considers these underlying factors, however, AI is still aiding patients and doctors. So it will be interesting to see how far AI progresses in the medical field.