The early detection of SPL by AI
The system achieves an accuracy of some 90.3% in detecting melanoma using smartphone images.
One of the most dangerous and most feared types of cancer is melanoma. This type of malignant tumour is responsible for over 70% of global cancer deaths. Over the years, professonals have based their diagnosis for detecting incidences of cancer on visual inspection. To do this, they have localised the SPL, Suspicious Pigmented Lesions, which if identified early, can be treated in primary care settings and can improve the diagnosis of melanoma and significantly reduce recovery time.
Today, technology is a great ally for the identification of cancer and other diseases. However, there are other types of wearables that help to identify the health status of their users. By doing so, many chronic patients can use monitoring and analysis to improve their health status and opt for telemedicine.
The difficulty lies in detecting and prioritising SPL quickly, as there are a large number of pigmented lesions that require analysis. With this in mind, researchers are working to quickly and reliably interpret and detect skin blemishes and rule out malignant tumours. A group of scientists at the Massachusetts Institute of Technology have trained a machine learning model that uses broad field images from smartphones and cameras.
How does it work?
The broad field image taken on a mobile phone shows large sections of a patient’s skin within a primary care setting. The automated system detects, extracts and analyses observable Suspected Pigmented Lesions (SPL) and, through a Deep Convolutional Neural Network (DCNN), determines the status of these blemishes. To do this, it marks them with different colours according to their status, thereby making it possible for the user to fully understand the results.
- Yellow = Additional inspection is required
- Red = Requires additional inspection or referral to a dermatologist
The functions extracted are used to better evaluate the pigmented lesions and to show the results in a colour map format.
“Early SPL detection can save lives. However, there is still a lack of capacity within the current medical system to provide complete to scale examinations of the skin”.
Luis R. Soenksen, a post-doctorate and expert in medical devices.
To achieve their results, the team of researchers used Artificial Intelligence, adding to the software 20,388 broad field images of 133 patients at the Gregorio Marañón Hospital in Madrid, as well as public images from the data banks. All the images were taken with ordinary cameras, that is, they were easily avalable to consumers. However, the dermatologists were responsible for visually classifying and comparing the lesions in the images.
The result of the research was that the system achieved an accuracy of over 90.3% in distinguishing the SPL of non-suspicious lesions, skin and complex backgrounds by doing away with the need for cumbersome individual images which take up a lot of time.
“We hope that our research revitalises the desire to offer more efficient dermatological examinations in primary care settings in order to generate appropriate referrals”
According to researchers, this method enables faster and more accurate evaluations of SPL and could lead to earlier treatment of melanoma.