Artificial intelligence systems could one day pose a large benefit to improving the accuracy and speed of medical diagnostic. Before the clinician can start to use the power of AI, the AI systems must first sort through x-rays with teaching algorithms so that they know exactly what they’re looking for in a diagnostic procedure.
Engineers today are using machine learning to build computer-generated x-rays that can populate new AI training sets. As the AI systems continue to sort through results, it’s possible that clinicians can then use the power of the artificial intelligence to interpret x-rays almost immediately.
AI could prove extremely useful in identifying a number of rare pathologies in medical imaging. With computer-generated X rays now being produced in almost limitless supply, large databases can be built-up to train new neural networks to identify various rare conditions within x-rays.
The team that’s developing these x-rays is using a technique called deep convolutional generative adversarial network or DCGAN. The x-rays are made up from one network that generates the images and another that works to incorporate real images, and since then it images to build a pool of accurate data for a larger augmented data set.
The new AI systems that were given the automated dataset were able to improve their accuracy by 20% for recognizing common conditions and up to 40% with the use of the synthesized x-rays for recognizing rarer conditions.
One of the best parts about using these synthetic x-rays is that researchers can conduct machine learning for AI outside of hospital premises and without violating any medical privacy concerns.
Of course, this is just one hurdle in the idea of AI being able to take on complex medical tasks with accuracy. As this technology continues to develop we may be able to classify x-ray images with a high degree of precision beyond the degree that is a human doctor could.