Artificial intelligence detects skin cancer more reliably than doctors

Artificial Intelligence

Artificial intelligence: As part of a study, both dermatologists and an algorithm should detect if pictures show black skin cancer or a benign birthmark. Of the 157 doctors, only seven people performed better than artificial intelligence. Nevertheless, the software can not replace the clinical diagnostics for the time being because differential diagnoses have not yet taken place.

Artificial intelligence is also becoming increasingly important in medicine. A study by the National Center for Tumor Diseases (NCT) in Heidelberg, published in the journal European Journal of Cancer , has now examined whether dermatologists or a specially trained algorithm works more reliably in detecting skin cancer . For this purpose, 157 dermatologists from twelve German university hospitals and the Algorithm 100 pictures were submitted, in which it should be judged whether the picture shows black skin cancer or a benign birthmark.

Artificial Intelligence
Artificial intelligence detects skin cancer more reliably than doctors

The algorithm, which was developed in cooperation between the German Cancer Research Center (DKFZ), Heidelberg University Hospital and the NCT, was able to achieve a better result on average than the clinical diagnosis by dermatologists. As Jochen Sven Utikal, Head of the Clinical Cooperation Unit of the DKFZ and Professor of the University Hospital Mannheim explains, the algorithm, despite its better performance, can not replace the diagnosis by doctors in a timely manner , but only supplement it. The scientists hope to be able to carry out more precise and quicker examinations through the parallel diagnosis by experienced doctors and software recognition.

Differential diagnoses are missing

Alexander Enk, director of the Department of Dermatology Heidelberg adds that “the algorithm of the medical practice can not do justice.” This is because he works very well, however, so far only recognizes black skin cancer. An experienced physician may be subject to the algorithm to detect this type of cancer, but may have more than a hundred differential diagnoses. In addition, some types of skin cancer are very rare and therefore hard to recognize by algorithms, as the necessary training data is not available in sufficient quantity. As Enk explains, “some types of cancer can hardly be diagnosed on a single image, but need more information, which a specialist receives, for example, by feeling”.

12,378 pictures as training data of Artificial intelligence

In the run-up to the study, 12,378 images were submitted to the algorithm, which served as machine learning training data. Subsequently, 100 images were submitted to the physicians and the KI solution, of which 20 images showed melanoma (black skin cancer) and 80 images benign moles. The dermatologists from the university hospitals of Berlin, Bonn, Erlangen, Essen, Hamburg, Heidelberg, Kiel, Magdeburg, Mannheim, Munich, Regensburg, and W├╝rzburg should then use the images to determine whether they would advise a patient to make an exact biopsy of a tissue sample or if they assume that the picture shows no skin cancer. Also the algorithm has got the same task.

Of 157 medical specialists, only seven physicians achieved a better result than the software, 14 of the dermatologists were on par and 136 of the doctors were subject to the algorithm. The performance of the doctors was independent of their position and experience. Both assistant physicians and specialists and senior physicians averaged a worse result than the algorithm.

The study now published is part of the Skin Classification project of the Federal Ministry of Health. Currently around 3,000 people die of black skin cancer each year in Germany. The new cases are in Germany per year at about 21,000 people. The main cause of black skin cancer is too high UV exposure of the skin.

If you like this post please share it to Facebook and like our Facebook page.

Next Article:Iran government kills more than 1000 protesters : United States