Machine Learning (ML) has evolved from a fringe technology to something more mainstream that is on most investors or companies lips. It is easy to dismiss this as technology born out of Silicon Valley that has no real-world application. This is no longer the case. ML has now moved beyond the research lab, to capture the attention of business leaders as they increasingly grasp the immense potential of Artificial Intelligence technologies to improve productivity and competitiveness. The ability to apply big data sets to ML is allowing us to fully harness the potential of this form of Artificial Intelligence in a way that was previously not possible.
So what exactly is ML? Wikipedia defines “Machine learning as a field of computer science that uses statistical techniques to give computer systems the ability to “learn” (e.g., progressively improve performance on a specific task) with data, without being explicitly programmed”. Put another way it is a type of Artificial Intelligence giving the computer the ability to learn without being explicitly programmed. If you feed a machine algorithm sufficient data, it will find the relevant patterns for you, and teach itself which ones are important clinically.
One of these advancements is the use of ML technology to obtain more information from medical images. Medical imaging is the application of science to visually represent the anatomy and biology of a human. In most instances, the interpretations of medical data is being done by a medical expert, such as a radiologist, and this will likely remain the case in the future. Advances in imaging technology in conjunction with ML capabilities are leading to a rapid rise their use in tasks such as risk assessment, detection, diagnosis and prognosis. ML is been looked at as a technique for recognising patterns that can be applied to medical images. Two of its major strengths is how it can help reduce human labour and lower costs.
A word of caution; it remains early days for the widespread adoption of this technology and, whilst a powerful tool, it can be misapplied. Deep learning algorithms for image recognition need to be trained on already labelled data where a patient has been given a definitive prognosis or diagnosis. The ML algorithm needs to be able to identify the combination of these image features for classifying the image. One of the pitfalls of ML is the effectiveness and accuracy of identifying a prognosis and the risk of misdiagnosis from data mining.
Computer-aided detection and diagnosis performed using machine learning algorithms can help physicians interpret medical imaging findings and reduce interpretation times. Provided we understand the strengths and weaknesses of this technology, we can potentially significantly improve patient outcomes, whilst reducing the overall cost burden for a health care system.