Launch HN: Darmiyan (YC S17) – Early Detection of Alzheimer's Disease
I'm told that launching on HN should come with the backstory of how we came to work on this, so I need to tell you about my grandmother, the most precious gift in my life. She was a poet who raised me, and was always full of life and stories to keep me amazed and excited. As the first female bank executive in a conservative society in the middle east, she was also socially progressive and outstanding. A brilliant brain. A beautiful mind. A few months before she died, on a sunny day, she told me: “Do you know what I want the most from my life?” I stared at her in silence. She continued: “To die decently while I still remember myself, my memories and my loved ones. It feels like as I’m getting older, I’m somehow losing my brain. As if my brain was lemon juice before and now it’s becoming lemonade.” That statement has been stuck in my brain ever since. Her wish never came true. She died not remembering even basic things of her amazing life.
Now, 14 years later, it’s been exactly 14 years that I’m researching human brain structure and function, and modeling how they degrade with age and by diseases such as Alzheimer’s disease. Believe it or not, it has been 110 years since the initial description of this devastating disease by Dr. Alois Alzheimer. Yet there hasn’t been much progress in pre-symptomatic diagnosis of the disease and no progress in finding a cure for it despite all advances in science and technology. One in every five Medicare dollars is spent on Alzheimer’s disease and the entire health care system will go bankrupt if no revolution happens in the field. “So, what is missing?”, I always asked. And what can be done to find the missing piece? I always tried to answer. Driven by these questions, I spent several years in biological physics master’s and PhD programs and neuroscience postdoctoral research.
Now I’m the founding CEO of Darmiyan. At Darmiyan we detect Alzheimer’s disease up to 15 years before symptoms, meaning exactly when treatments are feasible and brain damage could be slowed down just by simple life style changes such as regular exercise, eating well, and sleeping well. We do this early detection non-invasively, using only standard brain MRI. We have spent the last three years in Darmiyan developing and validating a software platform that models the human brain and simulates the tissue architecture underlying every individual voxel (3D pixel) of the brain MRI. Our proprietary methodology and results have been officially reviewed and approved by clinical Alzheimer’s experts at Stanford and the world leading Alzheimer’s expert and Nobel prize winner Paul Greengard. The most challenging part of our journey so far has been to get access to the largest MRI databases for Alzheimer’s disease and clinically validate the software. Now we have analyzed more than 3000 brain scans and our software’s predictions are 90% accurate.
My co-founders are Thomas Liebmann, PhD, a top-notch experimental neuroscientist who has managed to visualize the most hidden parts of the human brain through the eyes of the most advanced microscopes; and Kaveh Vejdani, MD, an extraordinary physician who always seeks complex problems at the interface of physics, biology and medicine and solves them with high level of knowledge, creativity and innovation. Thomas was my first office-mate in Stockholm 12 years ago. We met when I had just started my PhD at the Royal Institute of Technology (KTH) in Sweden and we became friends on the first day. Kaveh and I met at a classical music event in New York 7 years ago and have been close friends ever since.
Our vision in Darmiyan is to help all those people in the world who suffer from complex brain diseases such as Alzheimer’s disease. We want them to be diagnosed early and get c...
72 comments
[ 1143 ms ] story [ 1964 ms ] threadWhat is the current state-of-the-art in treatment (2017, Aug)? As an interested person, I have recently heard on Radiolab that MIT has done research with using 40Hz light to trigger brain-cleansing circuits that pulsate through the plaque buildups. Is this the most promising form of treatment at this time?
Another question I have had for a long time but have had no strong-science-strong-medicine-strong-research friends to ask in this domain is: What is the plaque? What causes it, how do we uncause it / remove the root of it, and is it a natural part of the brain chemistry out of balance, or a foreign invader?
Thanks a lot, hope these questions are within what you are able to talk about at this time. Wishing you the best of luck and success in your endeavors, and I look forward to living in an Alzheimer's-free world one day!
If the 15-year risk for the population taking the test is 10%, an unbiased 10% error rate would mean:
81% true negatives, 9% false positives, 9% true positives, 1% true negatives.
> Even with a false positive, the changes you would make (lifestyle changes, brain stimulation, continuous monitoring) would only be beneficial
Not if prioritizing them crowded out things you could be doing to address a real risk that is deprioritized because of the false result.
And, also, of course, an error rate doesn't have to be unbiased. With the same population, a test with these results would also be 90% accurate:
90% true negatives, 10% false negatives.
A 2 year multidomain intervention of diet, exercise, cognitive training, and vascular risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): a randomised controlled trial
http://www.thelancet.com/journals/lancet/article/PIIS0140-67...
What was the state of the alzheimer-detection-world prior to this?
I'm curious what you're comparing your results against to determine their accuracy. I thought there was no way to get a confirmed diagnosis of Alzheimer's short of an autopsy. (Obviously I'm no expert -- just going by what I've read.)
For example, did you do any hand tuning of thresholds? Did you do this tuning while looking at the scans, or did you design it without looking at ANY of the mentioned 3000 scans - and are you are basing your 90% accuracy on diagnosing a sample which is how representative of the population at large?
Is there a further description of this 90% statistic - do you have precision / recall and some sort of description of possible biases in this 3000 sample set? Based on another answer, it sounds like half are alzheimer's and half are 'similar' but non-alzheimer's. I don't want to be flippant, but I am always sceptical of 'accuracy' in medical diagnostics where a simple diagnosis of 'false' is 95% accurate.
Edit: I see you answered this second section in another comment, nevermind.
You all, are truly inspiring and dedicating your life to making a difference. It has truly been a pleasure to read about what you're doing, and I can't wait to see how you change the world. Thank you and your team, sincerely. and all the best!
Arguably, there will be no viable treatment unless asymptomatic volunteers get tested with our software so they can enroll in a clinical trail, so why defer the test?
That combined with the relatively high false positives makes this a pretty bad investment so far, in my opinion.
This is really a case where the rate of false positives and negatives (and the ages of the study population) with comparison to incidence is needed, and not just a % accuracy. Just unconditionally saying “no” gives you a pretty high prediction accuracy (83% lifetime accuracy for women at 65, 91% for men at the same age, based on remaining lifetime risk of developing the disease.)
Can you clarify how you reached the 90% number given that the detection is 15 years before the disease and your startup is only 3 years old. How were these numbers measured?
If I may throw in another question, what is the biggest challenge your company faces? It looks like there is no reason not to choose your product.
Regarding your second question/comment, we have just launched, and you are absolutely right, there is no reason not to choose our product. The biggest challenge is the huge skepticism in the medical community towards any novel technology related to Alzheimer's disease, given all prior failed big claims by others.
"At present, there is no definitive evidence to support that any particular measure is effective in preventing AD"[1].
"Now we have analyzed more than 3000 brain scans and our software’s predictions are 90% accurate." (original post)
"In the United States, Alzheimer prevalence was estimated to be 1.6% in 2000 both overall and in the 65–74 age group"[2]
I'm assuming the 3000 brain scans you're referring to is from individuals which progressed to Alzheimers, or at least a dataset with such individuals highly represented (if this were 3000 random individuals, at a 1.6% prevalence, that amounts to 48 individuals who eventually got Alzheimers).
So according to my calculation of Bayesian probability, with a 90% sensitivity (as I'm interpreting your comment), and a 1.6% prevalence in the population, a randomly screened individual with a positive test will only actually have a 12.8% chance of getting Alzheimers. So you'll be diagnosing lots of people so that they can have an impending Alzheimer's diagnosis hanging over their head for the remainder of their life without actually being able to do anything about it, and of this cohort just over 1 in 10 people will actually end up getting Alzheimers.
Please tell me you're only planning on offering this for researchers, and not actually going to try to get individuals screened? Or am i missing something about your value proposition?
Edit: fixed sensitivity vs. specificity errorI have never seen anyone over 75 that does not have some kind of dementia. Zero. Alzheimers and Dementia will soon be one of the same classification. It's happening.
Nature is just telling us, you have to go. It will be incurable, it's painless. She's being nice to us. Yes treat it, of course, but we have to let go sometimes.
PS, I work with seniors. People have NO CLUE to what these people go through. Zero.
And NO one can face this question, eventually, we die. Know the millennials don't believe that. They can't even comprehend it death. But it's true. Really. And it's OK.
So live life as it should be. For ALL our days are numbered. Don't worry, be happy.
:-)
- not the same as dementia, and never will be [1]. Dementia is an umbrella term that includes many diseases (e.g.: Parkinson's, etc), whereas Alzheimer's is just one of the many specific diseases with its own causes, symptoms, and chances of developing a cure.
- not painless. Apart from increased physical pain sensitivity [2], there's endless emotional pain that impacts the patient and their loved ones [3].
The point you're making about the inevitability of death is valid, but it shouldn't lead to the conclusion that we should just accept Alzheimer's as an incurable disease - of all the ways that a loved one can pass away, many families I work with would agree that this may be one of the worst.
Having worked with thousands of end-of-life clients, I am surprised that the research funding for Alzheimer's is far lower than that for cancer, heart disease, and HIV/Aids. My assumption is that people's view of Alzheimer's is rooted in the old and incorrect perception that it's an inevitable and natural part of aging, sometimes referred to as "senility."
Today we know a lot more about Alzheimer's, and I would challenge anyone to point out why Alzheimer's is fundamentally incurable - it's just a matter of when, and I certainly hope we'll find a cure sooner than later.
[1] https://www.kindlycare.com/dementia-vs-alzheimers/ [2] http://www.psychiatryadvisor.com/neurocognitive-disorders/al... [3] https://www.kindlycare.com/still-alice-portrait-of-a-disease... [4] http://www.aarp.org/health/brain-health/info-2015/alzheimers...
> PS, I work with seniors.
My fiancee says something similar, but she's a resident in a hospital, so she tends to tag "... in the hospital" or "... in the ICU" on the end of that statement because she's not seeing a random sample of all >75-year-olds.
There is currently no other method to detect and quantify micro-structural abnormality in the brain at presymptomatic stages of dementia, which is one of the main reasons why all clinical trials of Alzheimer's test drugs keep failing one after another.
When a disease-modifying treatment is found using our technology (and practically impossible without it), the person who knows the status of their brain health will be the one to benefit most from the treatment before developing symptoms.
How will your tech enable a disease-modifying treatment? Is the hypothesis your tech is the first way we can even measure?
Are you using standard MRIs and just a layer of software, or do you require some sort of special physical device?
This is all curiosity btw; I love the idea. I used to work on software for confocal microscopy and protein melting curves (to be clear, two different labs) before selling out and working on ads.
Darmiyan folks, is this correct? If so, what are the implications for the utility of the test?
(No matter the answer it feels like this research is important -- thank you!)
> Roughly 97% sensitivity (3% false negative) and 85% specificity (15% false positive)
So with this information, the calcuation of Bayesian probability is as follows:
So a 9.5% chance of actually getting Alzheimers within the following 15 years if Darmiyan's test is positive. I don't have much formal statistical training, so I'm all ears if I'm making a mistake in the calculation here.They also write the following:
> False positive here is not real false positive, as the software is detecting abnormality in people who are still cognitively normal.
Which is a terrible excuse - abnormality is only clinically relevant if it leads to disease. By this logic I can create the world's greatest test for cancer simply by saying that every given individual has cancer (100% sensitivity), and if they don't have it yet, they do carry the genetic abnormalities that will lead to cancer eventually - no "real false positives", right? Simply saying that "sooner or later, our test will prove correct" is not good enough here.
Founders of Darmyian: I commend your efforts in this space as a tool for research, and can potentially be very valuable for clinical trials. However, offering it to consumers as a screening tool when there are no proven preventative measures is - in my opinion - completely unethical and comes across as an attempt at trying to profit off fear mongering.
Zola, S.M., Manzanares, C.M, Clopton, P., Lah, J.J., and Levey, A.I. (2013). Am J Alzheimers Dis Other Demen. 28(2): 179–184. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3670591/
There is currently no other method to detect and quantify micro-structural abnormality in the brain at presymptomatic stages of dementia, which is one of the main reasons why all clinical trials of Alzheimer's test drugs keep failing, one after another.
You can think of Darmiyan's product as a quantitative virtual microscope that, among other things, can help pharmaceutical companies come up with a disease-modifying treatment by testing the test drug on "cognitively healthy" volunteers who could potentially benefit from the drug, if it's really effective, before it's too late. Right now there is no other tool in the market to identify and monitor microscopic abnormality in cognitively healthy brains.
I'm assuming the 3000 brain scans you're referring to is from individuals which progressed to Alzheimers, or at least a dataset with such individuals highly represented (if this were 3000 random individuals, at a 1.6% prevalence, that amounts to 48 individuals who eventually got Alzheimers). So according to my calculation of Bayesian probability, with a 90% sensitivity (as I'm interpreting your comment), and a 1.6% prevalence in the population, a randomly screened individual with a positive test will only actually have a 12.8% chance of getting Alzheimers. So you'll be diagnosing lots of people so that they can have an impending Alzheimer's diagnosis hanging over their head for the remainder of their life without actually being able to do anything about it, and of this cohort just over 1 in 10 people will actually end up getting Alzheimers. Please tell me you're only planning on offering this for researchers, and not actually going to try to get individuals screened? Or am i missing something about your value proposition?
I am working in a similar domain (www.16bit.ai) and am a radiologist. I think there is utility in this quantitative analysis despite there not being an intervention yet, despite what others are saying. The reason is because in order for others to measure the effectiveness of new treatments a gold standard and reproducible way of measuring the disease is necessary. I think drug companies working in this area would be quite excited about this if it can be shown to be a reliable predictor. As for direct to consumer marketing (ie. 23andme model) that some have suggested - its possible but it will be tough to educate patients to go and get their MRI images from their hospital for this purpose. I live and practice in Canada so my views and experiences of this may be a bit skewed. Best of luck!
"Our proprietary methodology and results have been officially reviewed and approved by clinical Alzheimer’s experts at Stanford..."
What is an "official review and approval" process?
"The most challenging part of our journey so far has been to get access to the largest MRI databases for Alzheimer’s disease..."
Which dataset is this?