Buzz
Although screening for skin cancer isn't perfect, a global team of researchers believes AI can enhance its accuracy. In a study published in the Annals of Oncology, they report that a deep learning convolutional neural network (CNN) trained on vast datasets can identify skin cancer more accurately than human dermatologists.
The team trained the CNN using over 100,000 images of malignant melanomas and benign moles. Co-author Holger Haenssle, senior managing physician at the University of Heidelberg, described the CNN as functioning like a child's brain, meaning it improves its ability to learn and refine its performance as more data is provided.
After training, the AI was tested on a new set of images it hadn't seen before. It correctly identified skin cancer from the images 95 percent of the time. In comparison, 58 dermatologists only detected 86.6 percent of malignant melanomas. Additionally, the CNN was less prone to incorrectly diagnosing benign moles as cancerous.
These results don’t necessarily indicate that AI robots will soon replace human doctors (or even pigeons) for cancer screenings. Instead, the researchers envision the AI program as a complement to dermatologists, potentially helping by analyzing images already stored in doctors' databases and offering 'expert opinions' on the likelihood of cancer.
Even as a doctor's assistant, the CNN still has room for improvement. The images it analyzed were predominantly of white patients and did not cover the full spectrum of skin lesions. It also faces challenges in diagnosing melanomas on fingers, toes, and scalps when using an image-based system. However, the researchers are optimistic that these issues won't hinder AI from playing a significant role in future cancer screenings. 'Given exponential development of imaging technology, we envision that automated diagnosis will soon transform the diagnostic process in dermatology,' the researchers stated.
