Skin cancer is the most common cancer and most people diagnosed with skin cancer first seek primary care. Primary care providers face the difficult task of distinguishing between common benign skin lesions and rare skin cancers.
“There’s been a lot of hype lately about the potential of artificial intelligence and machine learning (AI/ML) algorithms to help diagnose skin cancer,” says Owain, a clinical research fellow in the University’s Department of Public Health and Primary Care. Mr Jones says of Cambridge. “In primary care, AI/ML algorithms could make a big difference. More accurate assessment of skin lesions could lead to earlier diagnosis of skin cancer, improve patient outcomes and potentially increase survival.” there is.”
The burden on specialized dermatological services may also be reduced, and patients can be reassured when lesions are shown to be less likely to be cancerous.
Guided by this idea, Jones and a team of researchers from the University of Cambridge decided to conduct a study to establish an overview of the research state of AI/ML algorithms and to assess the currently available evidence for their efficacy and safety. Did.
“While we were aware of the emergence of commercially developed AI/ML algorithms and techniques to aid in diagnosing skin cancer, the evidence for the safety and effectiveness of these techniques remains limited. was not generally available,” Jones said.
Regarding studies, the researchers carefully considered which types of studies to include in the systematic review.
“We were interested in applying AI/ML algorithms to the diagnosis of skin cancer in primary care, but our initial survey found that no studies had developed AI/ML algorithms in primary care settings. I understand,” Jones said. “So we decided to develop an AI/ML algorithm to include all studies that could potentially be used in primary care settings. but meant that the review included a very broad overview of all available evidence.”
Of the 272 studies included in the review, none used primary care data and only two studies used data from populations considered similar to primary care clinical populations.
The data showed that AI/ML algorithms demonstrate promise for skin cancer diagnostic accuracy in the laboratory. However, Jones pointed out that there is a lack of evidence regarding AI/ML algorithm implementation and accuracy in real-world clinical settings.
“Our primary interest in this review is in primary care settings, and because of the current lack of evidence in settings with low skin cancer prevalence, broad adoption into primary care practice is unlikely at this time.” It turned out not to be recommended,” he said. “Our review also highlighted concerns about the datasets used to develop many AI/ML algorithms. It also highlights concerns about whether technology will keep us from being plagued by the prejudices of certain minorities.”
The apparent dearth of patients from black and minority backgrounds in the datasets used to develop the AI/ML algorithms is a surprising result, and was not accepted at the time the researchers began their review. “I didn’t expect it,” Jones shared.
“Another surprising result is the lack of real clinical practice and research, especially given the enormous amount of research that has been done in this area over the past few years,” he said. Stated. “This indicates that AI/ML technology aimed at diagnosing skin cancer is probably at an earlier stage of development than we expected when we started this research.”
The study authors conclude that AI/ML algorithms have great potential to support clinicians in the accurate detection of skin cancer in primary care settings. However, this review indicates that research in this area is at an early stage of development, and that the diagnostic performance demonstrated in the included studies may be more pronounced among populations with low prevalence of skin cancer, or There are concerns about whether it will be maintained in a dermatoscopic or non-dermatoscopic environment. Poor quality image.
“These algorithms are accurate, effective, cost-effective, and safe enough for clinical use, and increased access to skin lesion assessment increases the biopsy burden for specialized healthcare providers. It needs to be carefully evaluated to make sure it doesn’t cause cancer or contribute to the risk of skin lesions,” said Jones.
For future research, the researchers created a checklist highlighting important aspects that AI/ML developers should consider in developing these algorithms.
“We hope that the results of this systematic review, combined with this checklist, will help improve the quality of research in this area and help develop implementable techniques that will bring clinical benefits to patients and clinicians.” said Mr Jones. “Following this study, we are now evaluating the risks and benefits of using AI/ML technology to assist in skin cancer diagnosis in primary care settings by educating patients, the public, and health professionals. , is working on a qualitative study that evaluates the opinions of data scientists.”
