Improved AI tool shows high sensitivity rates in skin cancer detection

Ongoing learning of an artificial intelligence (AI) tool markedly enhanced the detection rates of skin malignancies. Compared to the first version of this tool, the sensitivity for all skin cancers rose from 83.3% to 99.5% in version 3.

Dr Kashini Andrew (University Hospitals Birmingham NHS Foundation Trust, UK) and his fellow researchers developed AI technology for the detection of skin cancer [1]. They now presented the 2022–2023 software version’s results in comparison to earlier findings of the first version that was tested in 2020–2021.

Over 2.6 years, a total of 22,356 patients were evaluated by the AI tool. This was triggered by a suspicion of a skin malignancy by the general practitioner. AI categorisation of the lesions was reviewed by a dermatologist outside the hospital in case of a benign result and by a university hospital consultant when cancer was suspected.

Between the first and the latest version of the detection software, further training of the AI with images of confirmed diagnoses took place. Sensitivity for melanoma recognition increased from 86% in 2020–2021 to 100% (59 out of 59 cases) in AI version 3. The latest AI version correctly labelled 99.5% of skin cancers (189 out of 190 cases), while the earliest software version achieved a sensitivity of 83.8%.

"This study has demonstrated how AI is rapidly improving and learning, with the high accuracy directly attributable to improvements in AI training techniques and the quality of data used to train the AI,” Dr Andrew commented on the improvement in detection. For pre-malignant lesions, the current sensitivity was also higher than in AI version 1, with a current rate of 92.5%. The AI demonstrated a specificity of 75.3% in recognising benign lesions. Notably, there was 1 case of basal cell carcinoma misclassified as benign by the AI, and the dermatologist had a benign-to-cancer overturn rate of 0.1%.

"We would like to stress that AI should not be used as a stand-alone tool in skin cancer detection and that AI is not a substitute for consultant dermatologists," study co-author Dr Irshad Zaki (University Hospitals Birmingham NHS Foundation Trust, UK) said. The use of AI version 3 was calculated to have saved 1,000 in-person appointments in secondary care. “Our data shows the great promise of AI in future provision of healthcare,” Dr Andrew concluded.

1. 
   
   1. Andrew K, et al. Continued Improvement of Artificial Intelligence in Identifying Skin Cancer. P1005, EADV Congress 2023, 11–14 October, Berlin, Germany.

 

Copyright ©2023 Medicom Medical Publishers



Posted on 28 November, 202328 November, 2023