November 10, 2022
Polish startup BrainScan has deployed an AI-powered system that it said can analyze CT scans and other medical images to spot brain lesions in a matter of minutes. It aims to cover all brain pathologies, which the company said its competitors do not do. The results of BrainScan’s AI system will support the findings of radiologists, whose ranks are declining as the workforce ages.
Felix Beacher, head of healthcare technology at Informa Tech, recently interviewed BrainScan’s CEO Szymon Korzekwa to talk about AI’s impact on and implications for the medical imaging market.
What follows is an edited version of that conversation.
Felix Beacher: Could you describe the system you have developed?
Szymon Korzekwa: BrainScan is an AI-powered software solution designed for CT scan analysis and auto-detection of brain lesions. It aims at improving diagnostics’ accuracy and efficiency by reducing under-reporting and assisting radiologists with patient prioritization.
The series of scans from the CT machine is sent as DICOM (Digital Imaging and Communications in Medicine standard) files to the PACS (Picture Archiving and Communication System) server, which is the computing device for securely storing, retrieving, managing, and accessing medical imaging information. It is a centralized device allowing data access requests to multiple other devices within the network.
Then the script filters all brain CT scans, anonymizes them and forwards them to the BrainScan Cloud. Algorithms powered by artificial intelligence automatically analyze scans finding potential pathological changes.
The results are saved in a form of infographic and are visible by the doctors as the next series in the study and saved in a DICOM file. The results (also stored) in a structured text form are returned to the PACS server from which they were sent. The advantages of a form of infographic is that it is available in every DICOM viewer used by doctors. And the whole process takes around five minutes.
The collection of imaging examinations from brain computed tomography collected by the company includes several hundred thousand unique examinations.
Currently we recognize a dozen or so pathological changes, but regularly update both the number of recognized lesions and the effectiveness of their detection, and, most importantly, indicate the areas of the brain in which they occur.
This number of examinations is a huge amount of information to be processed; it must be remembered that each CT scan consists of successive slices, also the number of unique images that we process is the product of the number of examinations and the average number of slices.
To illustrate the scale − if we assume that the average number of slices is 50, we are currently analyzing about 15 million images. Each of these thousands of imaging studies has a text description that must also be interpreted using text analysis methods.
Beacher: Could you describe a typical patient and what happens in the old situation?
Korzekwa: There are two areas and two types of patients in which our solution helps to improve patient diagnostics. One is Emergency Department patients that need urgent CT scan and quick analytics to know what kind of damage has been made to the brain.
In case they have a hemorrhage, it is crucial to quickly find out what kind of hemorrhage it is and how it needs to be cured. In this case, each minute is important and can change the outcome and − literally speaking − save the life of a patient. So in this case we facilitate the triage. The meaning of triage is “to sort or sift.” In the medical sense, triage is a system of collecting patient information and prioritizing patient care.
The second type of patients are those that are clients of teleradiology companies. In this case, we increase the efficiency of the work of the radiologist who describes the imaging examination. With the use of our software, it improves diagnostics’ accuracy and efficiency − the system is able to detect very small changes that are easily overlooked by a human, especially in the sixth or seventh hour of his work. The system, detecting all pathologies, will give the radiologist a second opinion with a list of detected changes, which gives greater certainty and precision.
In the old situation, the analysis of the CT brain scan took a very long time as the workforce shortage of radiologists constantly increases. This results in the fact that patients need to wait long and radiologists are stressed out not being able to fulfill all the hospital and patients needs. Sometimes it also happens that the radiologist is not available and this causes a lot of pressure for the whole team.
Beacher: Now let’s imagine the same patient is treated with imaging AI system in place. What is the difference?
Korzekwa: The most important factor is the time, and also the most crucial for the patients. The entire procedure of transferring data from and to the hospital and its analysis takes less time than the time it takes for the patient to return from the CT room to the emergency department, which is a huge time saver, innovation, and at the end of the day, a life saver. The whole system is about the patient and how to be able to help more people in less time, not making mistakes. This is where AI helps.
Beacher: Could you summarize the benefits of these kinds of AI-based workflow solutions?
Korzekwa: Currently, the methods broadly called AI work very well mainly in the areas of classification. Basically speaking, features like the number of colors, the number of shades of gray, multidimensionality, and a multitude of sections, are not a barrier for them.
For example, the human eye theoretically distinguishes between 500 shades of gray, while current devices return an image whose elements have from 4,000 to over 65,000 values. While our brain probably does not compete with digital methods and is still, as a small biological creature, capable of abstract thinking, our receptors, such as sight, smell, and touch are often not so effective.
All this also helps with the workflow, as the computer is able to analyze thousands of datasets in minutes, recognizing things for which we need more time and focus. This is crucial in emergency situations.
Beacher: What are the risks of the new system?
Korzekwa: People are sometimes scared of AI solutions and I think this is because they do not fully understand them − we have, as humans, always been resistant to innovations. However, I am convinced that systems such as BrainScan will become tools commonly used in diagnostic imaging in the next few years. Just like more than 10 to 15 years ago, the fluorescent lamp backlighting the X-ray examination was replaced by the monitor and more perfect software allowing the manipulation of imaging examinations, obtained with more modern methods of imaging diagnostics, such as CT, MRI, PET-CT, PET-MR, etc.
The technological details of systems that work and give benefits to facilitate or improve work are not as important as the values and help they give us.
Currently the systems are so advanced that there are not many risks − it is mostly in our head that we are not used to being diagnosed by the machine and we feel uncomfortable about that.
In terms of ethics, the issues are similar to other AI systems: If the machine makes a mistake, who is responsible, could we sue the machine and take the case to court? Still in Europe and the whole world the legislation does not follow technological development and there are a lot of doubts about how to put those policies into practice.
Beacher: How competitive is this space?
Korzekwa: A lot of companies are working on AI-powered imaging systems, but each concentrates on various body parts. In terms of brain hemorrhage, BrainScan is one of the leaders and we have probably the biggest dataset in the world on which we teach our AI systems and data is the key in such solutions.
Our system is also very accurate and our aim is to cover all the brain pathologies, which none of the (other) companies do. In general imaging, AI is rather dominated by startups who cooperate with big companies on their solutions, which is great as this topic needs a lot of iterations and agility and innovation and this is what startups are known for.
Beacher: In your experience, how are imaging AI systems being received by radiologists and radiographers?
Korzekwa: Similarly to the patients, radiologists sometimes treat it as unneeded innovation and have their doubts but later on when they try the system and find out that it is a huge support to their work and not a replacement of what they do, they are all into it.
Younger generations seem to be more open to innovative solutions but it all is really about education and showing possibilities. Probably it is similar with all humans − we are not so eager to try new devices but once we start, it is later hard to move forward without them. And with AI being more a part of our life in all areas, it is also easier to persuade the radiologists to make it part of their jobs.
The BrainScan system should be treated as the next generation tool provided to the doctor. After all, some of us still remember how the X-ray machine was the basis of imaging diagnostics, and the photo taken on the film was viewed by a radiologist attached to a negatoscope.
Then came the era of computed tomography and magnetic resonances. Digitally saved images gave room for software development, which enables image processing to extract as much diagnostic information as possible from the examination. It offers the doctor many options to change display parameters, image reconstruction, filtering and many more − that was the second step.
Now it's time for the next step, introducing further improvements, which are possible thanks to knowledge and technology, and which always help the patient in the end.
Beacher: What is your perspective on how medical AI systems are regulated in Europe and beyond?
Korzekwa: In my opinion, the AI systems are well-regulated in terms of being used by radiologists but as I mentioned before there are still a lot of ethical issues around them in case a system makes a mistake and (who takes) the responsibility for it.
But then, on the other hand, we are sure that the mistake made by humans supported by a system is far less likely than a mistake of a human without support especially after being overloaded and working several hours.
Maybe this is the way we should think about that − more of a support than looking for the responsible ones. If the legislation does not follow the development and (maybe even) stop it, we will never be able to acquire innovations and make such solutions help millions of people who need it.
About the Author(s)
Practice Lead, Healthcare Technology
You May Also Like