Man vs. machine – the benefits of AI in imaging

Radiology is a field that produces large volumes of data, which can no longer be managed without the help of intelligent systems. This is especially true when it comes to the interpretation of medical images. While this takes physicians years of training and experience, several hours of work and the highest level of concentration, artificial intelligence only requires a few seconds to accomplish the same task.

According to estimates, the amount of medical data is expected to increase by nearly 50 percent annually. Most of it - 90 percent – is generated in medical imaging. In 2019 alone, radiology will generate 675 billion gigabytes of image data. This translates into a staggering 13.5 trillion cross-sectional images – a quantity that's hard to imagine, let alone use. In fact, only 7 percent of these images are being processed. The 100th German X-ray Congress in May discussed these numbers. It is impossible to manually tackle this job, making machine support crucial to keep up with the ever-increasing demand.

Artificial Intelligence is not only able to work at consistently high speeds and precision, but it also identifies hidden patterns in data, giving physicians valuable support in diagnostic and treatment decisions. Machine learning is especially beneficial where diagnostics data has been examined by doctors and digitized.

Since April, the University Hospital Jena is the first site worldwide to use GE Healthcare's Deep Health Image Reconstruction (DLIR) in routine clinical applications. The system is able to reduce image noise in CT data, resulting in outstanding image quality. In an interview, Felix Güttler, Commercial and Technical Director of the Institute of Diagnostic and Interventional Radiology at the University Hospital Jena explains how DLIR works: "The system's artificial technology uses a so-called deep neural network (DNN). AI systems that are approved as medical devices typically no longer learn during their application. An algorithm is generated from the trained DNN. The artificial neurons are trained for a particular application and then 'frozen' for use." This ensures that the behavior of the algorithm is accurate and reproducible.

In addition to GE, other large manufacturers are also providing AI-based solutions for routine clinical practice – including Siemens Healthineers and its AI-Rad Companion Chest CT. In an interview, Ivo Driesser, global product marketing manager said: "The device is designed to help physicians make a faster, more accurate diagnosis by replacing manual image interpretation with AI analysis. The AI-Rad Companion Chest CT is an AI-based software assistant for CT images of the thorax." The algorithms can identify the heart, aorta, lungs and spine in the scans, combine several images into a 3D image and measure organs.

The Ingenia Elition 3.0T X MRI machine by Philips promises a safe and speedy analysis with the help of AI. Samsung provides AI technology and evaluation algorithms for both the GC85A digital radiography and the RS85 ultrasound systems.

Algorithms are able to deliver results in a fraction of a second and can be used cost-effectively anywhere in the world. In other words, they greatly benefit the overall healthcare system. "The bottom line for patients is simplified processes, quicker personalized care and more time with the attending physician," Güttler sums up the extra benefits of AI solutions for patients.

Having said that, machine learning will not replace us humans in the near future. That's because Artificial Intelligence is still man-made, relies on and is only as good as the quality of data it is trained with. What's more, medical professionals still have a difficult time understanding how AI actually makes an assessment. Right now, decision-making is basically seen as a kind of black box. Meanwhile, the algorithm should have a capacity for self-criticism and be able to express uncertainty and indicate unanswered questions. After all, this is the level of transparency that we expect from human experts and that ultimately builds trust. Researchers are already working on making machine learning more explainable, thus allowing humans to be able to control, correct and improve it. A machine must not be allowed to make a definite diagnosis by itself. "Simply put, I think when it comes to the patient's welfare, diagnostic and treatment decisions will always be in the hands of human beings," Felix Güttler feels confident. Ivo Driesser agrees and adds, "Humans will always have the last word. Software assistants are only a physician's companion."

In the coming years, the number of patients will continue to increase. That's why the diagnostic process and all hospital procedures must be optimized to save time and relieve undue medical staff burden. "Think of using Artificial Intelligence to manage hospital bed occupancy rates or to automatically match disease patterns involving hundreds or thousands of data points. Imagine a fully automated scheduling system or the chance to let AI take over routine measurements and monitoring," Güttler points out additional AI applications in medicine.

It's impossible to look at Artificial Intelligence without looking at big data at the same time. They go hand in hand. Wherever AI comes into play, thousands of datasets have been collected in the run-up. And vice versa: radiology, cardiology or pathology exams, as well as patients and their smartwatches constantly generate and store data. "The goal is to take advantage of all this data," says Driesser and points to another application of Artificial Intelligence: a digital twin. "Once a physician chooses a treatment, it could first be virtually tested. This allows an analysis of data and monitoring of systems to head off problems before they even occur, thus minimizing risk." Not only would the collected data be used effectively, but it would also realize the promise of personalized medicine. "The more I know about a patient, the more I can predict a patient's future," explains Driesser.