BillionPhotos.com - Fotolia

Artificial intelligence identifies skin cancer in lab test more effectively than doctors

Artificial intelligence has outperformed experienced dermatologists in identifying skin cancer samples, further highlighting the potential of the technology in healthcare

Artificial intelligence (AI) can more accurately identify cancerous skin cells than experienced doctors, research from the European Society for Medical Oncology (ESMO) shows.

As detailed in a recent edition of the Annals of Oncology journal, the scientists trained the AI’s neural network using 100,000 images of melanoma cancer cells, and normal skin birthmarks or moles.

It was then tested using 300 dermoscopic images, with 100 of the most difficult photos from this sample handed to 58 dermatologists to assess, so conclusions could be compared with the results from the AI.  

The doctors, more than half of which had more than five years’ experience, had to diagnose the potential cancer and say what type of treatment was required, if any at all.

While they correctly classified 87% of the melanomas and 71% of the birthmarks, the AI identified 95% of the cancer samples correctly.

The doctors then reassessed the same cases four weeks later and – after receiving further information about the patient’s age and gender – correctly identified 89% of the cancer samples and 76% of the non-cancerous samples.

Researcher Holger Haenssle said it is notable the AI performed better than the dermatologists, despite only having access to the image data.  

“The convolutional neural network [CNN] missed fewer melanomas, meaning it had a higher sensitivity than the dermatologists, and it misdiagnosed fewer benign moles as malignant melanoma, which means it had a higher specificity, and this would result in fewer unnecessary surgeries,” he said.

“These findings show that deep learning convolutional neural networks are capable of out-performing dermatologists, including extensively trained experts, in the task of detecting melanomas.”  

Based on these results, there is potential for the technology to provide supplementary support to doctors in the future to inform their clinical decision-making processes, added Haenssle.

“Most dermatologists already use digital demoscopy systems to image and store lesions for documentation and follow up. The CNN can then easily and rapidly evaluate the stored image for an ‘expert opinion’ on the probability of melanoma.”

Earlier in May, UK prime minister Theresa May talked about the importance AI could play in this field, saying it could diagnose up to 50,000 cancer cases and save 22,000 lives per year by 2033. She said continued improvement in these technologies will give the country “a new weapon in our armoury in the fight against disease”.

“Achieving this mission will not only save thousands of lives, it will incubate a whole new industry around AI in healthcare, creating high-skilled science jobs across the country,” she added.

The government is already acting on this theory by investing £75m to develop new centres, which will use technologies such as AI to offer better disease diagnosis. This comes as part of a wider £300m fund to push technology and innovation forward in the UK.

Experts also told the House of Lords Committee in November 2017 that there is a “huge opportunity” to use AI to identify cancerous cells for NHS patients. The biggest challenge facing the system, however, is organising and managing the large amount of data from across different health trusts.

Read more on artificial intelligence in healthcare

Read more on Artificial intelligence, automation and robotics

Join the conversation

2 comments

Send me notifications when other members comment.

Please create a username to comment.

Back in 1982 the "expert systems" of the day were already better than most GPs doctors in spotting early stage cancers - although it should be said that neither was picking up more than 50% because neither had current knowledge of what to look for.  
Cancel
Despite current AI has their bases on research works carried out some tens years ago, ICT has evolved in a  manner that accuracy in sensing has been improved highly. Moreover, those "expert systems" are now part of complex intelligence networks. On the other hand, a very high number of samples are provided for machine learning processes from the big data paradigm so it is possible to correlate variables that have been never taken in mind in order to accurate on diagnostics.
Cancel

-ADS BY GOOGLE

SearchCIO

SearchSecurity

SearchNetworking

SearchDataCenter

SearchDataManagement

Close