Artificial Intelligence

DERM is different to other AI

The machine learning algorithm built by Skin Analytics, Deep Ensemble for the Recognition of Malignancy (DERM), recognises the most common malignant, pre-malignant, and benign skin lesions. This includes melanoma, the most dangerous of the common skin cancers.

Almost all existing solutions for the recognition of skin cancer in images take a similar approach: using an existing ‘pre-trained’ neural network, and retraining it using skin lesion data. These ‘pre-trained’ neural networks (such as Google’s Inception network and Microsoft Research’s ResNet) are designed to perform very different tasks, such as large-scale image recognition (classification of 1000s of image categories, such as cats, dogs, and lamp-posts). While use of this approach gives a reasonable performance for an initial proof-of-concept, a solution of this type is inadequate for deployment in a medical device.

At Skin Analytics we’ve taken a different approach, designing all aspects of our machine learning architecture from the ground up for the specific problem we are trying to solve. This includes specifically tailored machine learning architectures, training methodology, and data augmentation for detection of skin cancer. Award

What is Artificial Intelligence? What is Machine Learning?

Artificial intelligence (AI) is the field of computer science which involves intelligence possessed by machines. Machine learning is a subset of AI that involves algorithms which are able to learn directly from data, without being explicitly programmed by a human. This technology is characterised by algorithms which are capable of learning complex representations directly from data itself. 

While in theory this gives machine learning algorithms the ability to learn concepts beyond the comprehension of the programmer that built them, until recently machine learning was far more limited. ‘Classical’ approaches depended on the programmer  creating code to extract useful ‘features’ from the data, so that they could be used by a machine learning algorithm. These ‘feature descriptors’ were limited by the perception and understanding of the programmer who built them, limiting the ability of the machine learning algorithm to learn from the data.

The implementation of ‘Deep Learning’ changed AI forever. Deep learning allows ‘features’ which describe the data to be learned directly from the data, without being explicitly coded by a programmer. In this way algorithms can be trained on data in an end-to-end manner, and highly complex features, beyond the understanding of the programmer, can be learned.

One of the biggest areas in which ‘Deep learning’ was first successfully applied was ‘computer vision’. This area of computer science involves using computers to understand high level concepts from images; essentially building software which can ‘see’. Given the highly visual nature of many aspects of medicine, deep learning is the perfect technology for application to assist in many medical fields.

DERM’s Performance

While machine learning and deep learning are incredibly powerful tools, they also come with risks. A key problem area in machine learning is ‘overfitting’, this occurs when a machine learning algorithm finds patterns in arbitrary noise in the data set, which negatively affects the algorithm’s ability to be useful in the real world or; generalise.. An example of this can be found in detection of malignant skin lesions in dermoscopic images. Often lesions selected for biopsy are marked with a blue circle drawn in marker around the lesion. If images of this type are used to train a machine learning algorithm to recognise malignant lesions, this feature could be incorrectly learned as an indicator of malignancy.

In practice, examples of overfitting can be far more abstract. This is especially apparent in medical applications where trends in data gathered from a particular site, demographic, or using particular equipment, can contain biases which negatively affect the ability of the algorithm to generalise when applied to new, unseen data. 

Observational research is extremely valuable to help prove the concept of your technology works. We have published an observational paper demonstrating DERM’s ability to identify melanoma within a dataset. Shortly we will publish an observational paper which further demonstrates DERM’s ability to find 11 different malignant, pre-malignant and benign conditions within a dataset. However, it is essential to validate any machine learning based medical device using a statistically powered prospective clinical study. This approach allows for an algorithm to be validated on a new, entirely unseen and representative data set. 

Skin Analytics conducted the world’s first ever statistically powered prospective clinical study assessing the ability of a machine learning based algorithm to predict the presence of melanoma in images of skin lesions. A paper containing the results of this study was published in JAMA Open in 2019, and showed that Skin Analytics’ computer vision technology was capable of detecting melanoma at the same level as that of a skin cancer specialist. 

Access to huge amounts of training data, processing power, and powerful algorithms does not equate to better outcomes. Without fully understanding the problem you’re trying to solve your returns will be diminished. Machine learning applied to healthcare is a complex multi-disciplinary area which requires expert knowledge in a number of fields. At Skin Analytics we have used this approach to build a powerful AI solution which is able to truly make a difference.

Partner with us


  • Our services are only available through our clinical partners in the UK. If you're worried about your skin and haven't been referred to us please speak to a doctor without delay.


Patient Safety and Quality Policy

Skin Analytics aims to provide best in class solutions for patients. We work to meet our customers and partners expectations by delivering on our brand vision with innovative and customer focused products and services.

We empower our teams to maintain a culture of innovation and continuous improvement, underpinned by an effective and efficient Quality Management System that complies with statutory and regulatory requirements.

We put safety and outcomes at the heart of what we do, to provide high quality products and services which are reliable, safe and compliant which is why we are so proud to have achieved ISO 13485:2016 compliance. MD 711728. 

Clinical Advisory Committee

Professor Scott Kitchnener

Professor Scott Kitchnener

Committee Chair. Primary Care Medical Service Director

Scott is a qualified specialist in medical leadership and management (FRACMA), a graduate of the AICD, a specialist public health physician (FAFPHM), while remaining a clinician with specialist standing as a general practitioner in rural practice. He has experience in health and education corporate governance and leadership in the private sector, public sector, higher education sector, international and Australian biotech sectors and in the military. His on-going research and experience have been in addressing health issues of developing nations, rural health and workforce, and particularly preventive approaches to public health challenges. 

Dr Niall Wilson

Dr Niall Wilson

Consultant Dermatologist

Niall qualified in Medicine from the University of Liverpool in 1991. He has been an NHS Consultant Dermatologist since 2000. He has special interests in skin cancer, hyperhidrosis and skin disorders in immunosuppressed patients. He has also been involved with postgraduate education for many years and is currently vice chair of the Specialist Advisory Committee for Dermatology, which oversees Dermatology training in the UK.

Dr Lucy Thomas

Dr Lucy Thomas

Consultant Dermatologist

Lucy is a Consultant Dermatologist at the Phoenix Hospital Group in London.

Professor Chris Hyde

Professor Chris Hyde

Health Economics

Chris is a professor of Public Health and Clinical Epidemiology at Exeter University. He is a School lead for research on test evaluation including systematic reviews, economic models and primary research. He leads the Exeter Test Group and is the diagnostics theme lead for PenCLAHRC. He is part of the Peninsula Technology Assessment Group (PenTAG). He directed the team delivering health technology assessments for national policy-making bodies, particularly NICE, from 2009 until 2015 and continues to support it by being the lead on HTAs of tests and through membership of its steering group. He is a long standing member of NICE's Diagnositc Advisory Committee and recently joined the National Screening Committee.

Justin M Ko, MD, MBA

Justin M Ko, MD, MBA


Dr. Ko joined Stanford Medicine in 2012 and serves as Director and Chief of Medical Dermatology for Stanford Health Care (SHC) while also spearheading the dermatology department's efforts around network development, digital health, quality/safety/performance improvement, and value-based care. He is active in a number of leadership roles within the organization including co-chairing the Clinic Advisory Council, a forum of medical and executive leaders of Stanford Health Care’s Ambulatory clinics, and as a Service Medical Director.

His passion for melanoma, early cancer detection, and improving care delivery drives his efforts and research around leveraging advances in machine learning and artifical intelligence to increase the breadth of populations that can be reached. He chairs the American Academy of Dermatology's Task Force Committee on Augmented Intelligence.