Deep learning in diagnosis: faster, clearer, safer

With 260 doctors, 570 beds, over 1000 births, 1,500 employees, around 25,000 admissions, 44,000 emergency cases and 68,000 polyclinic appointments every year, ASZ Aalst in Belgium is a busy hospital much like any other. Yet at the same time, it’s taking huge strides in diagnosis due to the efforts of three eminent doctors who have a vision for safer, more comfortable experiences for their patients through new technologies.

Concerns around the levels of radiation exposure through CT scanning are well-documented, but in terms of risk versus benefit, CT scanning almost eliminates the needs for exploratory surgeries – vastly speeding up the time to treatment. However, repeated exposure to radiation is something clinical professionals would prefer to avoid. Medical imaging specialist at Aalst’s General City Hospital, Dr Eddy Van Hedent and his colleagues Dr Laurence Trappeniers and Dr William Simoens, (Head of Radiology at ASZ Aalst), are looking to develop a screening programme for lung tumours that exposes patients to no more radiation than a traditional chest X-ray, with no impact on the clarity of image.

“The objective is to have the radiation dose for each and every examination score under the 25th percentile of the Belgian Federal Agency for Nuclear Control,” explains Dr Van Hedent. The most common CT examinations in ASZ Aalst are the thoracic and abdominal, during which “an annual exposure of 10 millisievert was hardly unusual.” Radiation in this area is measured inmillisieverts or ‘mSv’ and according to the International Atomic Energy Agency, most people are naturally exposed to around 2 mSv every year, just from their environment. But approaching the threshold of 50 mSv is a serious cause for concern, therefore, it’s of the utmost importance to keep exposure as low as possible – through faster, low radiation scans and limiting repeats.

Pioneering for Patients through Deep Learning

When ASZ Aalst invested in the Canon Medical Aquilion ONE GENESIS with AiCE, they took a significant step in achieving this vision and are rightly extremely proud to be a “European pioneer in AI-driven CT scans.” ‘AiCE’ stands for ‘Advanced Intelligent Clear-IQ Engine’ and is the first Deep Learning Reconstruction method using Artificial Intelligence to produce highly detailed images with extremely low noise. Deep Learning is an area of AI which uses layers of algorithms to process vast quantities of information independently. In the case of AiCE, high volumes of variable image data from previous CT scans ‘trained’ its algorithm to reconstruct CT images with the highest quality and at speed. So, it may come as no surprise to learn that it’s eight powerful graphics cards and AI application combined approaches the computing power of IBM’s Watson.

In the past all three doctors have experienced difficulties in properly visualising patients when CT scanning. The appendix, for example, is naturally situated in slightly fatty surroundings, which can make it difficult to locate right away. In the reconstruction of CT images, AiCE and the Aquilion ONE GENESIS can clearly distinguish various types of soft tissue, even at low radiation doses and without any contrast fluid (which can be swallowed or introduced intravenously into the body to allow the radiologist to distinguish normal from abnormal conditions) another important benefit of the technology.

Of course, the primary concern of Drs Trappeniers, Van Hedent and Simoens is the day-to-day reality of this technology for patients: a tremendous reduction in radiation doses, which is crucial for scans on children and those who regularly undergo CT.

As Dr Van Hedent explains, “We’ve now managed to reduce this [exposure to radiation] to an annual level of less than 5 mSv, because each thoracic and abdominal scan only produces about 1 mSv of radiation”.

The future

Of course, we are only at the beginning of understanding the true impact of AI in diagnosis. According to Dries Mahieu, CT/MR Product Specialist at Canon Medical Systems Belgium, work is underway on bringing AI into its Vitrea Advanced Visualization platform, for automatic detection of particular conditions in specific patients. Data collected from all scans will also give doctors a swift way to understand which patients should be prioritised in order of urgency. For Dr Trappeniers, AI holds great possibilities in the field of Computer Aided Diagnostics for CT and hopes to “be able to quickly and unambiguously determine whether there is air in the abdomen, for instance, or whether there is any internal bleeding or fat infiltration. That would enable us to make major progress, to the benefit of our patients.”

The continued development of Coronary CT scans is the focus area for Dr Van Hedent “Based on the results produced by the Aquilion ONE GENESIS, I expect that CT will soon surpass MR scans once and for all,” he states. “It’s only a matter of time before CT becomes the diagnostic tool of choice for coronary examinations. My experience with the Aquilion ONE GENESIS till now has shown simply that every single scan works, and it’s only going to get better.”

Progress is a two-way street

While the partnership and technology provided by Canon Medical is extremely important to the clinicians and patients of ASZ Aalst, the hospital is also a great source of experience, expertise and knowledge for Canon. “At ASZ Aalst, we have three extremely experienced CT specialists, who do not shy away from experimentation. We can learn a lot from them”, says Dries Mahieu. “For Canon Medical Systems Belgium, for all hospitals in Belgium, and even for current and future users all over Europe, ASZ Aalst is a reference centre. The doctors here know how to test the limits of the system and keep fine-tuning the applications we have developed day in, day out.”