Skin cancer is among the most common causes of cancer in humans. Last year in the US alone, over half a million people were diagnosed with this disease's variants. The usual diagnosis is made using clinical examinations and biopsies. Basal cell carcinoma (BCC), squamous cell carcinoma (SCC), and melanoma are the three most common skin cancer types. Actinic keratosis, nevus, dysplastic nevi are some examples of precancerous skin lesions.
Many of these lesions often tend to spread to deeper tissues too. Therefore, early detection is the key to effective treatment. Many dermatologists routinely perform skin cancer screening to identify premalignant skin lesions. However, the screening results are not always accurate due to the chances of over and under-diagnosis.
Scientists have found a more accurate way to identify skin cancer than the traditional way of diagnosis by dermatologists alone. They are using artificial intelligence (AI) to do so. Convolutional neural network (CNN) has proven to help reliably identify serious cancers like melanoma and nevi. In a recent study, Dr. Holger Haenssle, a senior physician at the University of Heidelberg, Germany, reports a surprising 'outperformance’ of CNN compared to 58 experienced dermatologists in identifying skin cancer following the examination of thousands of clinical images of the disease.
Several global startups are focusing on Big Data & Artificial Intelligence (AI)-powered technology to improve skin cancer diagnosis. British startup AMLo Biosciences – Protein Markers is one such startup that helps practitioners identify and treat a highly aggressive form of skin cancer called melanoma at its initial stage. It works by targeting specific protein markers in metastasized tumour cells to control the spread of this disease. AMBLor is a test that detects stage 1 and 2 melanoma by identifying protein markers, AMBRA1, and loricrin in the epidermis, and it has a short turnaround time.
French startup ANAPIX medical leverages ML algorithms to assess scan reports and patient data to help decide, thereby reducing the need for tissue biopsies. Its decision support algorithm DIAMELA helps the diagnosticians diagnose melanoma as it scans through the clinical pictures of the suspicious skin lesions. ANAPIX medical also has an algorithm called SKINAN that utilizes a database with details of multiple types of pathologies to identify melanoma-related anomalies. Its app SkinApp carries out diagnostic imaging of the skin lesions using a smartphone camera.
Similarly, UK-based Skin Analytics offers screening of melanotic lesions with the help of the algorithm Deep Ensemble for the Recognition of Malignancy (DERM) which identifies skin lesions with different malignancy potentials. Another US-based SkinIO helps people monitor their skin regularly with the aid of artificial intelligence and licensed dermatologists. Pictures of their mole, rash or bump on the skin are uploaded to a database through SkinIO’s app.
Often AI identifies atypical skin lesions based on the colour and distinct boundaries. MelaFind is one such technology being used at The Dermatology Specialists, New York. It uses infrared light to identify the pigmented lesions, and then AI algorithms analyze these images to diagnose skin cancer. SkinVision is an app developed by this institute that uses an AI algorithm to assess the pictures taken by smartphones of any skin lesion. MoleMapper is a smartphone application that helps you map and monitor the moles and bulges on the skin over time. Triage is also a smartphone app that leverages AI to screen skin disorders.
Canada-based MetaOptima improves the reliability of skin cancer diagnoses with the help of DermEngine, its software platform. It combines with a telepathology imaging system and artificial intelligence to assist dermatologists in detecting probable melanoma cases at an early stage. One of its previous innovations, MoleScope, has the possibility of an attachment that can convert an ordinary smartphone into a medical imaging tool to identify early-stage melanoma.
The recent breakthrough discovery by Scientist Adam Yala and his team at MIT was developing an algorithm that identified cancer with better accuracy. For this, they trained a model, 'Mirai’, with over 200,000 cases that contained several parameters for better clinical adoption and screening. Such advancements that enhance the diagnostics accuracy of the conventional models help scientists scale clinical adoption.
The AI-based diagnosis of skin cancer is a promising one, but research shows the presence of a considerable amount of margin of error in some cases. Therefore, the current state of technology still warrants the use of biopsy to arrive at a conclusive decision. Moreover, the existing AI technology has to integrate more diagnostic and prognostic factors to make AI more efficient. While making a clinical decision, it could take into account other similar pigmented lesions on the patient’s skin, history of exposure to skin cancer risk factors, duration since the present lesion had appeared etc.