Neural network able to diagnose skin cancer better than doctors

New York patients who are undergoing testing for potential skin cancer may be interested to learn that a deep learning convolutional neural network may be more accurate when detecting benign or malignant skin lesions than human professionals. The study to test this involved researchers from the U.S., France and Germany.

In the study, the convolutional neural network (CNN) was reportedly shown more than 100,000 images of both benign and malignant moles and skin cancers complete with a diagnosis attached to each image. It was noted that the images were dermoscopic, meaning each of the lesions were magnified 10-fold. The CNN reportedly improved its ability to differentiate between malignant and benign lesions with each image. Once the CNN was trained, images that it had never been shown before were introduced.

The results of the study showed that the CNN correctly detected 95 percent of melanomas while dermatologists correctly detected about 86.6 percent. The dermatologists' ability to correctly detect melanomas went up to 88.9 percent when they were provided with the patient's demographics. While researchers do not expect the CNN to replace professionals, it could potentially be used as a tool to improve the rate of correct diagnosis.

When it comes to detecting malignant melanomas or other potentially deadly illnesses and conditions, a doctor misdiagnosis can be devastating. Such an error could prevent early treatment that may reduce the impact of a condition. A delayed diagnosis could even cause a patient to require treatments that have more serious risks. If there is evidence that negligence resulted in the misdiagnosis or delayed diagnosis, an attorney may file a medical malpractice claim against the doctor and practice.