060-Dirk Smeets PhD

Robert Lufkin 0:02
Welcome back to the health longevity secrets show and I’m your host, Dr. Robert Lufkin. Magnetic resonance imaging can provide valuable information about hippocampal atrophy in people at risk for Alzheimer’s disease. When we combine Mr. with artificial intelligence to determine the brain volumes, the sensitivity and specificity of the findings increased dramatically. Dirk Smeets PhD is on the Faculty of Engineering Department of Electrical Engineering in the center of processing speech and images in the Medical Imaging Research Center at the University Hospitals as Gus Duisburg Lewin, Belgium. He is also Chief Technology Officer at isometrics Corporation.

And now please enjoy Dirks presentation

Dirk Smeets 0:57
from everybody to just talk as part of the Alzheimer prevention virtual conference.

My name is Dirk Smeets, and I am the Chief Technology Officer at psychometrics. And today, it’s a real pleasure for me to present you my perspective on the role of imaging AI in dementia. And today, I would like to focus on how it can enable a timely and accurate diagnosis of Alzheimer’s disease.

Today, it is estimated that about 50 million people across the globe are living with dementia. And by 2030, it is expected to grow to 82 million people posing an enormous burden on the health and social services. As you probably know, dementia is an umbrella term for a group of cognitive and behavioral symptoms severe enough to disrupt daily functioning. There are different types of dementia. About 60 to 70% of dementia cases are believed to be caused by Alzheimer’s disease. Making the timely and accurate diagnosis of Alzheimer’s disease is essential in the management of this impending public health threat. Unfortunately, there are struggles as well. About 50% of the primary care physicians are not up to date with the care in Alzheimer’s disease. And that leads also to more than three years of time between the first symptoms and the diagnosis. So the question we should ask ourselves is, how can we optimize the early detection and diagnosis of patients with dementia? For an answer to that question, let’s have a look at biomarkers for Alzheimer’s disease. Biomarkers or biological markers are measurable indicators for a certain condition, in this case, Alzheimer’s disease. For Alzheimer’s disease, we can extract biomarkers from magnetic resonance imaging in short, MRI from cerebral spinal fluids in short CSF or from Positron Emission Tomography or in short pet. Of course, when using those biomarkers and clinical routine, they should be affordable, available, preferably non invasive, diagnostic, and, if possible, predictive. When looking at all of these criteria, it seems there is an important role for MRI to extract biomarkers for the diagnosis of Alzheimer’s disease. Indeed, literature has demonstrated that MRI biomarkers allow the early detection of the disease that they can predict cognitive decline, and that they help with differential diagnosis. Moreover, they may predict other biomarkers like amyloid, which could be extracted from CSF or pet. But before diving in the possibilities of MRI biomarkers for Alzheimer’s disease. Let’s first have a look how MRI is used today in the clinical care path of patients with dementia. Today MRI is D primary imaging modality in the diagnostic process, as

was also clear from a survey that was performed in Europe, where over 70% of the centers indicated they use MRI as a primary imaging modality. Here in the US, the usage of MRI is expected to be even higher because of the wider availability of MRI. When looking at the role of MRI in the diagnostic process, we see two main reasons. The first one is the investigation of core pathologies, or the exclusion of alternative pathologies. The second role is the detection of pre dementia and dementia and the differential diagnosis of the types of dementia.

The investigation of the MRI scan is performed by a radiologist, typically, a radiologist is doing a structured visual evaluation. Thereby, he or she can use visual rating scales. The most frequently used scales or scores are the medial temporal atrophy, the MTA score and the physique score for vascular lesions. A higher physique score is seen in patients with vascular dementia. The second most common type of dementia, and a higher score on the MTA scale might be indicative for Alzheimer disease, as mentioned earlier, the most common type of dementia. According to the ESR survey that was conducted by Professor of Noy over 50% of the institution’s indicated they always use the MTI MTA scoring system for the assessment of the medial temporal atrophy also sometimes referred to as the hippocampal atrophy and almost 30% use it regularly. Less frequently used are the global cortical atrophy scale or the GCA scale and the coup dam scale for posterior atrophy.

Do use a visual rating scales leads to a better and more structured radiological report. But there are still some complications with those scales. Although most radiologists that participated this the survey that I mentioned earlier, are confident in using those visual rating scales. There is still a significant portion that was not confident and there are different and important reasons why they are not entirely confident. First, the scales made depends on the scan protocol. Also, a training is required in the usage of the scaling. We know also from literature that there is only a moderate to good agreement within and between raters. And also quite challenging is that those scores should be adapted for a patient age. What is the highest score might be normal for an old patient, but abnormal for a young patient? And maybe most limiting of at these visual scoring systems is the lack of civility. For example, the MTA scale ranges from zero to four with only discrete intervals. So, the question raises of how we can easily extract more detailed information from the MRI to be more sensitive to pick up pathological brain changes and as such, reduce the time to diagnosis. And this can be done through a quantitative MRI analysis. This is typically don’t on the T one way that scan, which is a scan that entails a good contrast between the white and the gray matter. The reason why this quantification of certain structures is a meaningful biomarker is illustrated on this slide on the left side. We know from analyzing 1000s of healthy control persons that our brain is shrink at the age of 30. This happens with about 0.1% per year, and this accelerates to 0.3 per year, percent per year at the age of 70 years. As a result of this normal aging process, a healthy person loses about 10% of his brain between the age of 30 and 75. For some neurodegenerative conditions, like Alzheimer’s disease, the brain shrinks faster. The atrophy can be between one and 2% per year. Moreover, it can occur assessment asymmetrically, for example in patients with them with semantic dementia, or regionally. Unfortunately, the survey that I mentioned earlier with the European Society of neuro radiology indicated as well that 76% of the centers never uses quantitative MRI analysis. In 80% of the cases, this is done for specific cases. And in 6%, it’s

always done. If they do if they did so. Most of the respondents were looking at hippocampal volume, because of the difficulty detect small differences, but also total brain volumes and lower volumes are frequently used because of the system’s sensitivity to take to detect neurodegeneration. The main hurdles identified by the respondents were delimited availability of software, the lack of time to use workstations, and difficulties into interpretation. So we need to think of how we can easily integrate quantitative MRI analysis in the clinical workflow and represent the results in an easy to interpret way. And in my opinion, this is where imaging AI comes into play.

Let’s have a bit of a look on the definitions. Actually, artificial intelligence or AI already exists for a very long time. The first day algorithms were developed in the early 50s. But over the year, years, AI has become more complex. In machine learning a subfield of AI algorithms are trained to solve tasks tasks, by pattern recognition, instead of specifically programming them how to solve the task, like it was done before. More recently, deep learning has entered the arena. Here, algorithms are trained with deep neural networks with multiple layers, and that allowed to learn relationships that are even difficult to understand for humans. And there is also a good reason why AI, in particular deep learning are becoming more and more popular nowadays. On the figure, you can see that since 2014, the AI performance is beating the human performance on imaging detection task. I need to stress here that this is not on medical imaging. But on normal imaging. I think personally that this is not yet the case for medical imaging, but it is also probably the situation that will come there as well. The artificial intelligence algorithms typically require heavy computations. And at isometrics. We have chosen to do those computations in servers that are located in the cloud. So images are sent directly from scanner or fax to the cloud where they’re analyzed and the results are sent back towards the pecs in which they are ready for the interpretation. As such, all results are nicely together with the original images available at the moment that the radiologist starts the reading. And this is an important prerequisite to have a I help with the quantitative MRI analysis for dementia. See more Integration AI should also not be a black box. For that reason, we also send back the quantitative MRI results together with segmentations. Those segmentations are color coded images as you can see here that show how the AI algorithm has interpreted the images and allows the radiologist to check if the algorithm was correct. Besides a seamless integration, and the necessary data to verify the results, the intuitive interpretation of the imaging AI is key to accomplish the translation of the AI results into faster and more certain clinical decisions. On this slide, you can see the reports of the AI based quantitative analysis with on the first page, the whole brain volume and the lateral ventricle volume and the hippocampal volumes and the inferior horns of the lateral ventricles. This first page is intended to detect abnormal atrophy early. The second page, which you can see on the right side of the slide, contains the quantification of cortical volumes and cortical asymmetries. And that should help to differentiate atrophy patterns to help with a differential diagnosis of dementia. In order to enable that timely diagnosis of dementia, supported by imaging, sensitive markers aren’t needed. If we have a look at the literature, we see that whole brain atrophy, as well as ventricular enlargement are sensitive markers for the progression of neurodegeneration.

However, these markers are not specific to Alzheimer’s disease, and therefore, the pattern of gray matter loss might be more disease specific. For example, the volume of hippocampal structures precedes the onset of symptoms by several year and the ratio there is a more sensitive marker than the individual volumes of the hippocampus and the inferior horn of the lateral ventricle. Structural MRI can also be used to differentiate Alzheimers disease from other pathologies, such as vascular dementia, frontal temporal lobe dementia, and dementia with Lewy bodies. For example, frontal temporal lobe dementia is associated with frontal and or temporal atrophy. Whereas for vascular dementia, the presence of diffuse white matter lesions lead to a global atrophy of the brain. In contrast, dementia with Lewy bodies usually does not show specific structural abnormalities. Furthermore, the importance of structural imaging is highlighted by the fact that according to the internationally accepted diagnostic criteria, such as those of the National Institute of Neurological Disorders, the diagnosis of vascular dementia is excluded when no vascular changes aren’t noted on the MRI.

So with the reports, and specifically the second page of the report of IQ brain, which is the name of the software, we have built a signature that fits those atrophied patterns, you can see three senators on the right on the right with on the left side to help the individual in the middle, a pattern of an Alzheimer disease patient with a more pronounced atrophy in the hippocampal area. And finally, a patient with frontal temporal lobe dementia with a more pronounced atrophy in the frontal and temporal cortex. I would also like to show you some exalt examples of how this quantification can support the interpretation. The first example is a quantification of a person of about 80 years old. And you can see that each of the quantification is in the green zone which allows the which shows that in comparison with the healthy population, it falls within the normal variation of that healthy population. That’s indicated that here, the whole brain is the 32nd percentile, the lateral ventricle at the 23rd percentile. So that’s within normal ranges. So, there is no suspicion of neurodegeneration here. The second example has lower whole brain volume, the second percentile, which can still be part of a normal variation, but it is on the low side. But what is also clear is that the lateral ventricles are above the 99 percentile, so not explainable anymore by the normal variation. So this is in line with dementia neurodegenerative process have taken as taken place here. In the third example, you can see that the hippocampus is normal, and the inferior horn of the lateral ventricle as well. So again, an example where we do not expect this person to be affected with Alzheimer’s disease. Example number four shows an example with a low hippocampal volume and two large inferior horn of the lateral ventricles, indicating a clear atrophy in the medial temporal lobe. And this is suggestive for Alzheimer’s disease. Then finally, as a last example, I would like to show a full example where the ratios of the lateral ventricle and the whole brain is too large. Also, the inferior horn of the lateral ventricle over hippocampus is quite large. So this, again, might be suggestive for Alzheimer’s disease. Although it’s also important to check for other potential forms of dementia. I would like to end with an clinical case of a 65 year old person with complaints of gradually progressive memory deficits. And because of the initial onomastic presentation, there was a suspicion of a prodromal, early onset Alzheimer’s disease. But when analyzing the MRI, there is a clear localized atrophy in the frontal lobe, thanks to the AI based volumetric analysis. And based on that, together, of course, with the clinical symptoms, the tentative tentative diagnosis of a behavioral variant of Frontotemporal dementia was made. So

this is an example where shows how the AI based fully metric quantification of the MRI image could support the differential diagnosis of that person with dementia. So I hope I have demonstrated how AI can help with the early detection of Alzheimer’s disease and to differential diagnosis of dementia today. But let’s also have a look on what the future could bring.

As mentioned earlier, the number of dementia patients is high and will unfortunately grow further. Luckily, with disease modifying treatments for Alzheimer’s disease being affected, as expected, there is a positive view on the future. However, these disease modifying treatments are likely to work only in an early stage of the disease when Alzheimer’s disease is not progressed too far to too far. And that would require the health care systems to diagnose Alzheimer’s disease earlier. And unfortunately, we also know that there is a delay of 18.6 months due to the limited healthcare capacity and with the even growing number, that means we need to act all together we need to act to try to detect patients earlier to diagnose them earlier and hopefully to treat them effectively with the expectations of treatment Coming. And for that, let’s keep our fingers crossed for the beginning of June when the FDA will share its opinion on other Kinomap. So when those are expected, we also need to think about the personalization of the treatment. Of course, that comes after the detection and the diagnosis of the disease. For the day for the detection, I also foresee in the future that the MRI based detection algorithms of AI should be combined weight, easier assessable metrics like for example, cognitive decline. And then finally, when the patient is diagnosed, it is important to stage the patient to know whether he can be selected for treatment. And then once selected for treatment, if there is also an important role for safety reading. And let’s dive a bit into each of these elements. First of all, I would like to present our cognitive testing tool that we’re currently developing, which is a companion DM, which includes validated questionnaires for cognition, fatigue, and symptom tracking, and also cognitive tests on processing speed, visual memory and digital span. And this is something that could be digital technology that could be used, together with the AI that can be applied on imaging, to be also as a kind of triage to help when the number of patients that needs to be diagnosed earlier is growing and growing. Once diagnosed, as mentioned, we need to stay patient. And for that reason, we are working on predictive models, in this case, event based models. With these event based models, we have shown that a mastic MCI patients, they score in generally higher than non domestic subjects, and that demonstrates the utility for the screening of patients that might be eligible for the treatment. Of course, further research needs to be done. But it is a very important role that we see here for artificial intelligence in this, in this case, combining clinical measures, with biomarkers from all different sources to find those patients that benefit most from the treatment that will hopefully be soon on the market.

And one when the treatment is on the market, and the patient is treated. It is of course also very important to monitor the patient, as we see from some of the current trials that there might be some amyloid related imaging abnormalities. And we distinguish two abnormalities, the hemorrhages and the edema. And with the ongoing trials for other Kinema which for which we hope there will be positive news in June for the FDA, there is in those trials, it has been shown that there is an occurrence of about 40% of amyloid related imaging abnormalities.

Those area related events can lead to clinical symptoms like worst case that but also dizziness, nausea, headaches, diarrhea, confusion, and visual disturbances. So altogether, we need to think about neuro imaging protocols for both area each and area. So with that, I would like to conclude my talk of today I see an important role of artificial intelligence in the personalization of the treatments for Alzheimer’s disease in the future. With artificial intelligence on imaging, already available today to see the invisible to help also with the standardized and quantitative reporting by quantifying atrophy patterns that are specific for diseases causing dementia. And when there is a big need for having more patients die diagnose. We need to complement those technologies with other digital tools like patient applications for a phone that could do cognitive testing and ask for validated questionnaires. And if we apply that well, all together, the health care system is hopefully able to identify prodromal patients where the patient app can create awareness can patients with gotten complaints and also raise campaigns on the importance of early diagnosis. And also, with the use of the more in hospital technology, we can reduce the time to affirm diagnosis by quantifying septal atrophy, by difference by differentiating with other diseases causing dementia, also, potentially helping the subtyping of Alzheimer disease to find those patients that will be most most benefiting from the treatment. And of course, as mentioned, there might also be a role in the post market studies of the treatment to also investigate potential safety issues with the treatment. I would also like to end with saying a little word on platform that we recently launched. That’s the alt imaging platform, where we try to prepare radiologist for a world with disease modifying treatments in Alzheimer’s disease. It is an educational platform for the management of Alzheimers disease. There will be CME credit modules in their educational chapters, interactive case studies and educational videos. It is completely unbranded, and it hoped it aims to train the whole healthcare system to be ready once the the treatments are available. And with that, I would really like to thank you for your attention, and I hope that you can enjoy the rest of this virtual conference. Thank you so much.

Unknown Speaker 32:21
No, this is not intended to be a substitute for professional medical advice, diagnosis or treatment. Always seek the advice of your physician or other qualified health provider with any questions you may have. Never disregard professional medical advice or delay seeking of it because of something you’ve seen here. If you find this to be a value of you, please hit that like button and subscribe and support the work we do on this channel. Also, we take your suggestions and advice very seriously. Please let us know what you’d like to see on this channel. Thanks for watching and hope to see you next time.