In the era of increasing demand for imaging tests as well as the amount of data and requirements for doctors, AI will allow for safer and higher quality work in the current environment. Today, medical everyday life is a heavy workload, time pressure, greater expectations of clinicians and patients, the need to make quick and, importantly, correct decisions, which are more and more often subject to verification, and finally common screening. All this in the face of the huge amount of medical data available for analysis.
Diagnostics imaging, next to radiotherapy and robotic surgery, is one of the areas where technology plays a key role and perhaps is the most influencing & beneficial of the final effect of a doctor’s effort.
A classic X-ray is one image… modern computed tomography contains thousands of them. It is estimated that in 1980 around 3 million computed tomography examinations were performed in the world, in 2015 over 80 million. In many areas of our lives we hear that we are not keeping up with technology… the limits of human capabilities often limit machines moreover often machines overtake people… the camera captures the image faster than the human eye … the computer analyses data faster than the human brain, which, nota bene, by itself has much greater abilities …
In these areas the dynamic development of technology will be visible, which in the era of new challenges will be… supporting the doctor!
There is no “I” in a team.
Artificial intelligence will not replace radiologists! They will be replaced by other radiologists who can maximize its benefits. The synergy of radiologists and AI algorithms based on deep learning will ensure the highest quality and accuracy of research results.
One of the DLAD (deep learning automatic detection) algorithms, by analyzing MR angiography images, independently detected brain aneurysms with a sensitivity of 91% –93%, and supporting the radiologist in the diagnosis, improved the probability of an aneurysm detection by 4.8% –13%. As a result, thanks to such cooperation, from 5 to 13 patients out of 100 respondents could be correctly diagnosed.
Radiologists reading CT scans of patients with suspected COVID19 with the help of AI significantly increased their efficiency- the accuracy increased from 85 to 90%. Not much? What if in those 5 better diagnosed cases out of 100 your child or parents were?
Less means more.
Next year, it will be 50 years since the first CT scan was performed. The pursuit of quality and extension of diagnostic possibilities has made one CT examination even take up several GB.
Thanks to the low-dose technology available, a CT scan generates only 20 times more radiation than a classic head X-ray (and several hundred times more photos!) and almost as much as you will probably take during one intercontinental flight. AI support in image processing and parameter adjustment will allow for a significant radiation reduction and safety improvement on an unprecedented scale. AI will advance image acquisition with all imaging techniques (CT, MR and PET / CT)
Today the AI algorithms have allowed to improve the quality of screening CT examinations of the chest with an unusually low dose of radiation, almost equal to the single X-ray image. Radiologists considered more than three times more scans performed with the support of AI as diagnostic images, i.e. those that allow them to assess the patient’s condition, compared to the scans not supported by the algorithm.
Stanford University researchers used algorithms to support the performance of PET/ CT examinations. Thanks to this, they were able to perform the scans, although so far only in laboratory conditions, using only 1% (!) of the current dose.
Such results will change the perception of the role of imaging tests in the treatment process. Safe examination, performed several times faster, with lower costs, allows to increase their frequency and thus improve supervision over the treatment process. Thanks to greater efficiency, it increases the availability of tests – with the same number of devices, more can be made, reducing the waiting lists.
Is medicine repeatable?
AI technology will play a significant role in the comparative analysis of research results. The ability to identify changes and progress over time is a major challenge, especially in the growing number of oncological, cardiovascular and neurological cases where regular examinations are becoming an integral part of “managing” the treatment process.
Early diagnosis of disease progression will result in faster adaptation of treatment and, consequently, improvement of its results and reduction of costs.
There are approximately 400,000 hip arthroplasty procedures performed in the US each year. Each patient undergoes annual check-ups. The results need to be compared with many previous studies. By analysing previous images, AI estimates the risk and informs the doctor about the need to take action, such as referral for further tests.
Another challenge is monitoring a patient with a chronic disease, e.g. cancer. If the radiologist has several disease areas under control, the algorithm automatically measures them and informs you which of them have changed and in what direction. Such information is an objective indication of the effectiveness of the treatment offered to the patient and the need for additional measures.
How to untie the Gordian knot?
All system members, including hardware and software manufacturers, are working to improve efficiency through standardization of naming, schemas, accurate prioritization, access to patient information and structured reporting. Modern integrated solutions should address the challenges of urgent cases, follow-up visits, radiologist schedules and the current increasing workload, also based on database meta-analysis. Research results indicate that some radiologists spend “as much as 35% of their day trying to prioritize reporting.”
On average, 46 seconds saved in accessing the integrated clinical data and image system may seem relatively insignificant at first glance, assuming that describing 20 examinations a day per year, the radiologist can save over… 60 hours!
Only the integration of EHR and PACS solutions in the field of deep learning and advanced analysis of data from disease history, drugs, laboratory test results, treatment plans in conjunction with data from diagnostic imaging will open up new opportunities and bring unprecedented quality and precision.
What if it could be done 10 times faster?
Magnetic resonance imaging is considered as a safe scan, therefore its number is dynamically growing. Unfortunately, they are also extremely time-consuming, currently taking from ten minutes to an hour, and they cannot be accelerated relatively easily- as in a CT- by increasing the number of detectors. Parallel imaging is a technique that speeds up scanning. However, so far the computer used only prior information (previous patient examination) or information obtained from a very small number of reference images.
In August 2018, Facebook and the New York University School of Medicine announced a collaboration using artificial intelligence to speed up the time needed to complete MRI scans by up to 10 times.
The currently tested artificial intelligence algorithm called DeepcomplexMRI enables parallel imaging based on the careful analysis of a large number of images used as data for training and creating artificial intelligence algorithms, and as a result can lead to a 10 times reduction in examination time.
AI algorithms support radiologists also at the stage of research descriptions, e.g. in performing automated segmentation of heart structures, it shortens this stage of work on the description of the examination from 238 s to 0.38 s (!).
Technological Esperanto
VNA is a new standard for storing images in a standardized format and interface, making them available to healthcare professionals no matter what camera produced the images. The systems we deal with today, proprietary to a specific manufacturer, dependent, closed with limited availability and introducing non-standard solutions will become less attractive to the entire radiological environment developing towards the growing role of artificial intelligence.
