skin cancer and AI

Finding the Mole: Using AI to Detect Skin Cancer

Skin cancer is the most common cancer among people in the United States. It is estimated that almost 10,000 individuals in the US are diagnosed with skin cancer every single day. If detected and treated early, almost all types of common skin cancer such as squamous cell and basal cell carcinomas, as well as melanomas, can be removed and are highly curable.

The problem is that detection is not always possible. Many patients may not notice subtle changes in their moles, many physicians don’t have time to examine a patient’s every mole during their annual physical, and not a lot of people consider booking an appointment with a dermatologist a priority. Thankfully, artificial intelligence may soon be able to help.

Skin Cancer and AI – What We Know So Far

In 2018, scientists published the results of a study comparing deep learning convolutional neural networks (CNN) with actual dermatologists when it came to detecting skin cancers. Although it had already been postulated that CNN could play a role in melanoma detection, the technology had never been compared directly to dermatologists. The results were promising – the Google Inception V4 CNN architecture could be trained to spot the difference between a melanoma and a harmless mole with a high level of accuracy. Higher, in fact, than most of the 58 dermatologists it was up against.

To conduct the study, researchers used magnified, high-resolution images of skin lesions. The images were analyzed by 58 dermatologists, who were asked to identify the lesions as either a melanoma or a benign growth. After their initial diagnosis, the dermatologists were provided with additional information, such as patient age, gender, and the region of the body that the image was taken from. From the initial image alone, dermatologists identified 86.6% of melanomas. After they received further information, this number rose to 88.9%

Staggeringly, the CNN technology correctly identified 95% of melanomas.

What Could This Mean for The Clinic?

Although this technology is not yet ready for clinical applications, it shows great promise when it comes to improving outcomes in patients with skin cancer. This kind of machine learning AI could offer expertise that matches a seasoned dermatologist, but wouldn’t require a visit to a specialist. Family practitioners could use this technology during routine annual physicals, assisting with early detection and in turn improving the clinical outcomes of skin cancer patients.

As well as AI technology, 2D and 3D skin imaging systems have recently been developed that allow doctors to take a closer look at skin tissue. These imaging systems give physicians the ability to characterize normal and abnormal cellular architecture, which isn’t usually possible with the naked eye. In a way, it’s like putting a patient’s skin under the microscope…but you don’t have to separate the patient from their skin first. This can be used to determine microscopic changes in the skin that may be indicative of skin cancer, without requiring an expensive or painful biopsy.

Caliber I.D. is one company providing physicians with an advanced look at their patient’s skin. This US based company is creating clinical imaging technologies that utilize confocal microscopy to provide a series of images which can then be stacked to give a 3D representation of the area of tissue. Their VIVASCOPE system offers a non-invasive, real-time look at a skin lesion which may or may not be cancerous, helping to determine whether further analysis (such as a biopsy or referral to a specialist) is necessary.

In future clinical applications, it is likely that this confocal microscopy approach will combine with the learning ability of the CNN technology to allow physicians to treat patients more effectively. With further development of the technology, AI and 3D imaging could be utilized to identify a majority of harmless skin lesions, leaving doctors and specialists with more time to focus on patients with more difficult or severe lesions.

Isn’t There an App for That?

It’s the year 2020, so naturally there are over 20 different apps available that relate to skin cancer detection. Most of these apps simply allow users to track changes in the shape or color of their skin lesion, but can still be beneficial by providing a reminder to perform a skin check at home, or suggesting scheduling a visit to a dermatologist. One app, however, takes things further and utilizes AI in skin cancer detection.

The SkinVision app allows you to take a photo of a spot on your skin, then harnesses an AI algorithm to provide you with a risk assessment. While it doesn’t offer a diagnosis, it instead classifies your skin lesion as either low risk, low risk with symptoms, or high risk, allowing you to decide how to proceed. While many users may prefer a “yes” or “no” answer rather than a scale of risk, this is a great example of the integration of AI into the medical field despite the fact that it still has limitations.

Early detection is a key factor when it comes to a positive skin cancer prognosis. The types of technologies described here could have a real and dramatic impact when applied to a clinical setting, potentially saving thousands of lives every year.