This calculation tells you your chances of being sick
HTML-код
- Опубликовано: 17 май 2024
- If you (or a friend) have endo symptoms:
Here’s my advice as a person who isn’t a doctor or an expert but has had some bad experiences with this condition.
1. It can be difficult to diagnose this condition partly because the symptoms are so different for everyone and partly because it’s hard to know what counts as “severe” pain. I thought the amount of pain I was having was normal for many years. If you have periods that are painful enough to be bothering you then it’s worth talking to a doctor.
2. GO TO A DOCTOR WHO SPECIALISES IN ENDO (once you’ve been diagnosed). Please please please don’t just go to a normal gynecologist. I did that and it was a huge mistake that cost thousands of dollars and caused damage that couldn’t even be fixed in my second operation. The surgery for endo is quite complicated and requires dedicated training to master, and yet the majority of doctors offering it haven’t done that training. For more details on how to find someone good I recommend that you:
4. Read some books about endo. In particular, I liked How to Endo by Bridget Hustwaite. It was approachable and full of useful advice. When I first got diagnosed I looked up endo online but that was pretty useless. It was only after reading this book that I got some idea of what was going on. That said, this condition is not well understood yet and everyone has their own methods of treating it that might not work for you, so don’t trust everything you read everywhere.
Here’s the diagnosis app I used (this isn’t an endorsement): ada.com/
If you want to learn about causal Bayesian Networks, The Book of Why by Mackenzie and Pearl is excellent.
Timestamps:
00:00 My medical story
02:48 Using Bayes' Rule for diagnosis
06:17 Causal Bayesian networks
12:34 Homework 
Наука
Hey Mithuna, I'm glad you are doing well again. Thanks robot overlords ;)
Thank god!
Thanks Dom :)
Praise be.
I'm so glad you figured this out and you're doing better! As someone with a chronic illness that no doctor can seem to figure out, I understand the dilemma.
Oh man, I'm sorry to hear about your situation.... It's frustrating when you *know* something is wrong but doctors say there's nothing serious. I really hope you find out what the issue is :(!
@Sean g 137 #NewPhysics shut up
@Sean g 137 #NewPhysics wtf is wrong with you man ???
You do seem to have a serious and chronic issue of Scientitis.
Joke aside I hope you will be able to determine what is wrong.
Me to.
I hope you are doing well, Mithuna! Wish you all the best
I hope my daughter (17) and her mother watch these special videos. Thank you for sharing your story. Woman are scientists.
Happy to hear that you're doing well! Great video too.
It's a great thing that you told this story! My wife went thru a similar, long, winding trip to a similar, blessed, resolution. I'm glad you're doing better, that you now have bright, clear, zooming future ahead - free of symptoms, uncertainty and undue pain. It's so wonderful that you've been able to fashion a blessing for others out of your response to a big challenge.
Glad to hear you are okay. My mom has dealt with this and the side effects from follow-up surgeries since the 80's. She's okay but to be honesty it hasn't been easy, if only medicine back then is where it is now.
I'm so sorry to hear that. It's a poorly understood disease even now, so I can't imagine how tough it would have been to have it in the 80s when the treatment and awareness was much lower. I hope she's ok ❤
Oh my! So glad you're recovering, Mithuna. Excellent video as usual!
We all wish that you are better now! And we all wish you the best!
From our hearts.
That is a scary story, very glad you are ok now! I agree a mixture of AI and Doctors advice seems the best way forward 👀
Thanks Josh :)! Something like that seems to make sense to me!
Bayes' Theorem is one of those phenomenons that is relatively simple, yet whenever I am to apply it, I always forget its existence.
It's so true, happens to all of us!
Omg! I’m glad you’re doing better. I hope you’re thriving ✨
I just found your channel via a comment you left on a Vlogbrothera video and I'm really glad I did! This seems like a very charming, clever and useful corner of RUclips. Thank you! :)
Woah ! Glad you are fine now. Wish you all the best😊
This is a lovely intro to Bayesian thinking! Thanks so much! Also glad to hear you are on the mend. I know first hand that recovery from invasive surgery can take a while and is PAINFUL! so good luck with all that.
Very glad to hear you're well now and the surgery was successful. Very interesting video about algorithmic diagnosis. I didn't realize it had progressed so much. :) I'd imagine doctors can work with it to get a lot of "second opinions" quickly and on the spot.
So happy you are okay now! You are a great person, Mithuna! Wish you all the best!
It will not replace doctors or most professionals but it will certainly enhance their performance.
I know I'll get downvoted for this, but I think you are wrong. Just like it was unimaginable to replace horses with cars, it is also unimaginable to replace doctors because they have been such a cornerstone of life on this planet for so long. But eventually AI will be able to perform the functions of a doctor better than humans. There is no law of the universe that says "doctors must exist". Once we have a solution that can cure disease far better than humans, there will be no more need to have human doctors. Just because things have been a certain way for our whole lives doesn't mean they must continue on this way.
@@mathoc5273 Whatever you call them, there will always need to be a human somewhere in the loop to take the corner cases and work with the AI both to teach it a new trick, if needed, or guide it, if it got something wrong. That person will need to be very well educated in medicine. If you don't want to call them "doctor", fine, but it will still essentially be what this person is, even if they may not even see patients directly in most cases.
@@rocketsurgeon2135 Exactly. What I do not see AI doing for a very long time is being 1) empathetic - part of the medical professions work is just being there for the patient or the family of the patient. 2) Second order (or even third order lying and lie recongnition). The patient is lying, the doctor knows that the patient is lying but lies to them, to get the nugget of truth - happens more often that anyone thinks. 3) Delivering the diagnosis. Most people who are for AI doctors do not get this for some reason. Patients have the psychological and social need to play the part of the "sick" (Parsons "sanctioned deviance") and to have a person to disagree with, even if they know the diagnosis is the correct one.
Generally Medicine is more like a craft or an art than a science and AI can be a valuable tool. I see AI replacing doctors the same moment when it replaces everything and everyone else.
The AI doctor results should be shown to the doctor after they have a candidate for the diagnosis. If you reveal what the AI suggests first, then there might be Bias in the diagnosis.
I think an AI should be used to verify or suggest alternatives just to make sure the doctor has considered all possibilities.
Great point!
Good idea! This should be true of every situation where multiple opinions are being incorporated. And ideally, when choosing one, the suggestor should be anonymized.
Yep there’s been studies on this actually and your conclusion is exactly right. The rate of false positives goes up quite a bit with suggestion, which makes sense. That said, it depends on what one’s stance is - there would be a golden intersection that sits between “what I haven’t thought about” and “what I should investigate for”, and that’s what you’d be trying to aim for.
In real life (as a GP registrar myself), people already come with feared diagnoses in hand a lot of the time. So then the history taking also is heavily influenced by that. In a perfect diagnostic world, people would have exactly the same symptoms for exactly the same problems and care about each problem equally…but humans are human and it’s more normal not to, meaning that the doctor’s daily battle is a battle against bias - a battle that’s hard because it’s cognitively difficult to process that bias in addition to already having to apply our own algorithm/Bayesian probabilities actively.
@@dabi_ Yes, well said David! I even found that with that app, the way I emphasise my symptoms makes a difference to my diagnosis. For example, it matters which symptom I say is my main concern, because it sends the app down a different line of questioning. It still usually comes back on track and identifies endo, but it's so interesting that even this model is so easily influenced by the patient's suggestions. I think this is part of what went wrong with all the many GP visits I'd had over the years (I've had many doctors because I moved so much). I thought period pain was normal (even though in hindsight crippling pain is probably never normal) and so though I'd mention it, I wouldn't make as big a deal out of it. I thought that the worse thing was the heavy blood loss, because that might be leading to low iron and therefore the fatigue. Most doctors agreed and sent me off with iron pills, even though my blood tests showed that my iron levels weren't bad. The power of suggestion. I guess.
IMO they should rename it to "statistical analysis" and not AI. But as a tool, it may be very very helpful and useful.
So happy you got your diagnosis and (hopefully) have started feeling better 😌💭☘️💝
I’m so glad you’re ok! 💜 it’s great to see you back making videos
Thank you so much! I'm feeling better now so I have plans to make a lot more!
This is an amazing video!! Thank you for posing these questions! I'm glad to hear you are recovering well from surgery!
Glad to know you're all right now. Take care man.
You've got one of my favorite channels, Mithuna! Plus, your story is pretty inspirational. I'm very glad you've recovered! 🙏
That's really kind of you, thank you :)!
Glad that you are doing well at the moment.
9:21 I think the main explanation would be that the difference in prevalence gives a factor of 3, while the difference in likelihood of symptoms only give a factor of 2, so the factor of prevalence becomes more important. If you have made the prevalence of A to be 1/3, then the difference in prevalence gives a factor of 2, and both probabilities would have been the same! The alternative change would be to make the prevalence of B to be 2/5, so that the difference in prevalence gives a factor of 2.
The factor of 3 is basically the proportion of people not having A vs having A, because there is a 50-50 chance of getting B.
I haven't done any of the math on paper yet, so I might be wrong, but that is my stab at the "explanation". Honestly, I don't think there is a more "intuitive" explanation (?), because humans are just bad at probabilities in general.
Nice seeing you here! Both of you have very good pedagogy :)
Btw, subbed to your channel.
I agree with your analysis. I made a throuough post with the math, summarizing what I did:
Basically, make a Punnett square for the probabilities (weights) then put them together. The "3" comes from the fact that (1-p)q is three times p (1-q). These are the only cases that do not occur in both A,B given tiredness.
So, although B is twice as likely as A, 2 is not the relevant scaling. This is because the probabilities given to us from your flow/causation diagram are conditional probabilities on only one of the diseases being present so you should not compare p and q but p (q-1) and (p-1)q.
This is the intuition.
This thread is awesome, thank you!!
Glad to hear all is well!
You are a real doctor! You’re not a medical doctor, but that doesn’t mean you’re not real! 🙂
I've always thought that health problems did nothing but slow down my studies and hinder me from getting closer to my dreams. But seeing you learning from your symptoms, I was encouraged very much. That's what real physicists are like, huh?
Glad to hear that you are doing well! Brilliant video 😄
Glad to see you back... Hope you are parfectly fine now. The video was great..👍👍
Glad to hear your ok. Turning your experience into a teaching moment, pretty nifty. The changing probability of B reminds me of the Monty Hall problem. Kind of an a-ha moment, although I'm not quite sure I really comprehend it entirely.
Good to know that you are getting better. Take care
Glad you are doing fine now. Very happy for you Mithuna❤️
We are so grateful that you are feeling better❤️💖✨🙏🏾
Thank you :)!
The world needs you. Thank you.
I have an answer , i wonderd that too about why a and b couldn’t equate , not enough room here, lol. I’ve forgotten more math than I currently remember, and it’s funny, how memory works. I taught myself maths but not out of books, just in thoughts I guess. Quantum physics was taught to me….by my 6th and seventh grade teacher mrs.Drake, who looks a lot like Joan Feynman, same mannerisms too. Life is stranger than fiction. Thank you for being you.
I'm sad that I only learned of this today. What a trooper you are! You completed a PhD in physics while dealing with this! I'm glad you're doing well now(or at least at the time of this video).
Your explanation made my day mithuna. Take care
I am so grateful that you are feeling better!
Thank you :) :)
Well done.. love your positive attitude
Glad you're alright/better now Mithuna! The first 20 seconds really had me worried!
Will definitely give your homework a look, although I should admit probability theory was not one of my fortes!
I’m really sorry to have scared you!
Probability is tricky for everyone, don’t worry. I hope looking at the homework might give you some new insights into probability though :)
wishing you a speedy recovery EDIT: you're already recovered... that was fast!
Glad to know you're okay.
I just downloaded the app; this is one initiative I'd love to help.
If you try it out, let me know what you think!
Glad you figured it out!
da faq dude your channel has been a part of my life for more than 6 years, and I gotto protest that you frightened me with this intro. Now, as a social scientist, I had developed an interest to phsyics and astronomy over the years, and you've had a huge part on that. Not that I'm invested by any means with your being or this channel, but it's still sad to hear that some person you like and admire might be no more sooner than expected. All that said, I'm most certainly glad to hear that you are alrite afterall. Sry for the broken Englando, I'm not really a native speaker.
I'm really really sorry to have scared you! I'm touched that you were so concerned though. Thank you for your kind message!
Free diagnosis..
That's really cool.
You can call it a pointer to a diagnosis
I came across your light video, then found this. I happen to be suffering from this right now so this was really helpful
Great video! I'm glad you are feeling better. I don't think computers should tell doctors what you have until after they have filed a diagnosis. This may help the computer come to a more accurate result, but it will also keep the doctor from making a biased decision.
Beautiful pictures 😅 happy to hear you're doing ok now :)
KEKW
Glad you finally got diagnosed and received treatment! Also, a Bayesian video 😍
Coming to why it does not balance, I think it is because the prior is not what you initially thought it would be.
In particular, B alone is 3 times more probable than A alone, because not having A is more likely than not having B.
If you assumed that A and B could not occur simultaneously (and thus it was A with 1/4, B with 1/2 and neither A nor B otherwise), then the posterior probabilities would be equal :)
That's such a good way to think of it! And straight from the Bayesian's mouth too, thanks Le :D
@@LookingGlassUniverse My pleasure :D
Glad to see you again ☺
I just wanna hug you right now and tell that you are so special and beautiful. Good to see you again 😊💖
Thank you Mithuna for this great video! I look up to your strength doing all your amazing work while struggling with such a difficult condition, and to your courage sharing something so personal! A few more points:
I think such an AI can only be as good as the data it's fed. Constructing the network is even harder than applying it, with noisy/fuzzy labels, small sample sizes, and limited opportunity of controlled experiments.
I'm so glad you're understanding of doctors who make mistakes. Like many other fields, medicine is now too expansive for any one person to understand. There's no shame in GPs / primary care doctors accepting algorithmic help. I think we still have to rely on human common sense for when a brittle AI... well... breaks.
Loved the "I'm not a 'real' doctor" one. You aren't the only PhD who feels that way :-)
Thank you Bela :)!
Yes, that's so true! Constructing this network is the bit that's really very hard, and I don't think we're there yet and may never be able to do it accurately. Like you pointed out, that's all the more reason for the human to be there to check it's not badly off base.
When I was in hospital every time a new nurse or doc would see my wrist ban they'd be like, "Ooh, what kind of doctor are you?" Every time I'd have to reply, "A doctor of mathematics." It got very embarrassing :P What kind of doctor are you?
@@LookingGlassUniverse Thanks for your reply! I'm a maths PhD too, but not nearly as smart as you :-)
I'm glad your doing well now.
Glad you are alright. I always enjoy your videos. I too had a medical issue that doctors couldn’t diagnose. I went to many different doctors and all I accumulated were bills from many tests and scans. The pain eventually went away and I feel better, but it lasted for a couple years. I am frustrated that I never found out what caused my pain or if it is something that will return. I will download the ada app and use it as an additional tool in the future. Thanks and I hope to see more videos from you soon! 👍🏻💜
It is so frustrating to be in that position, I'm so sorry that happened to you. I know the feeling. So many tests, so many bills, so few answers. I hope that whatever it was, you will be able to figure it out and it never comes back.
You find time to do this video in your schedule timetable. Great job. Take care
I literally said yayy and then realised you’re ill I’m so sorry
That's really cute!
A very interesting and well presented topic. Thank you very much :)
My favorite part of this video is that you have " Godel Escher Bach " in a stack of books.
Beautiful smile! Great you're doing well after all. Take care!
Ouch! I think of that as one of the most painful conditions a person can have, so sorry you've had to deal with that, and I hope your surgery helps a lot!
Things get even more complicated when the therapies or follow-up diagnostics have significant side effects. Of the people who receive a positive result, some proportion of the false positives can end up with bad outcomes because of unnecessary engagement with the medical system. I think that's a factor in mammography, for example.
Yes, you're absolutely right, there's more to consider than just the probabilities. We really need a model that takes into account how bad it is to be wrong. In the mammography example it could lead to a huge amount of suffering for no reason. In other examples, even if the probability of a condition isn't super high it makes sense to act out of an abundance of caution. There needs to be a way to factor that in..
I'm very sorry you've been unwell. I wish there were something I could do to help.
But, given the quality of those pictures, I'm not surprised they had difficulty with the diagnosis.
Shocking how bad their cameras are! It's a wonder they can treat the condition at all with equipment like that.
Glad to hear that you're recovering well now and that the surgery went really well for you.
I'm not doing the homework though. I already have Geography and Organic Chemistry courses to focus on, not to mention a job search to get back on track
Good luck with the job search! I give you permission to skip homework (from me) until you've finished with everything else on your plate.
@@LookingGlassUniverse Cheers Maithuna. Glad your hospital stay is now done I have once coming up soon.
Fair to say my plate is pretty full.
Really good to see you active on YT again. I'm glad to hear you are doing well.
I was wondering about some of your old quantum mechanics videos. They seem to be marked as private. Will you be reposting them or new videos covering the same material?
Also, do you think you might ever cover quantum electrodynamics/quantum field theory?
Thank you! Here's the playlist :) ruclips.net/p/PLg-OiIIbfPj2RNY2-tYO2JsR9uw7Rw22z
The qft books are on my table because I've been thinking about making some videos on it... let's see :)
@@LookingGlassUniverse unfortunately when I go to that playlist many of the videos show as private videos and I can't watch them 😭. Like the first 6 in the list are all private.
The best doctor I've ever talked to impresses me as being only as good a problem solver as the worst engineer I have ever known. Medical field is crap.
I completely agree with u. Great video.
Amazing video! Thank you.
Get well soon 👍🏾😎
Just subbed. Seems like a lovely space.
Glad to hear you are recovering well :). Speaking from experience, any kind of chronic illness is really draining, not just physically.
re: ML in healthcare: any kinda of automation suffers from automation bias, meaning, mistakes get amplified. This becomes crucial in a medical setting (or any place where it is affecting people directly, e.g. credit approval, aspects of fraud detection). So ML in healthcare is great when it is in a supporting role: triaging, or confirming a diagnosis. There's also some work on prognosis, treatment plans, and patient follow-up, but all of these should be in a supportive role, otherwise you are quite likely to make costly mistakes. I also like David Schumerth's comment about "blinding" the algorithmic results.
PS: I hate the term AI :-p, it's so vague and mostly used as marketing. And great video, as always you covered most of the challenges with relative ease :)
Great points raised! If we were using these systems for more than a supporting role then all the issues you mentioned are compounded by the fact that most of us view computer algorithms as objective and correct. You’re right that it’s better as an aid to human experts
It's fascinating that the go-to example that is used when teaching Bayes' theorem is medical diagnosis. Medical diagnosis is often unintuitive and cannot be performed successfully without a baseline understanding of statistics or even a full statistical analysis. It may not be viable to expect all doctors to be trained in advanced statistical methods on top of the million other things they have learn. I believe that software can go a long way toward bridging this gap.
Thanks you for sharing your experience.
There are several of my favourite books in your library, I bet you've carefully selected what goes into this pile :) I would love to see you talking about books that influenced you
Hey Mithuna, great that you finally got a diagnosis and that you are feeling well after the surgery.
I think there are at least two breakthroughs missing before we get AI doctors. One is inference in an opened system, which medicine functionally is, and the other is collecting appropriate and accurate information from patients. The latter seems to be closer to a solution considering systems like Watson, GPT-3; certainly more effort is put into this problem.
The current system assumes a closed system and lets the doctors create the questions which is fine for conditions that fit this scheme nicely but it could be prone to bias, e.g. looking for a disease when a condition is actually caused by inappropriate diet.
The good thing about having an closed system approximation is that it can be expanded and improved while people are working tackling the open version. Either way it looks promising. A serious push from a couple of smart people should be enough :)
Thanks :)!
Your analysis seems very interesting! I agree that these systems can ingrain biases, eg that everything can be solved by medicine as opposed to lifestyle. To be fair though, doctors seem to have a bunch of biases including this one. I can imagine new biases being introduced though.
Can you explain what an open system is?
Absolutely, doctors have these biases which the system will pick up. They will probably be a bit different so maybe they can balance each other out in some situations.
How I would define it is that an open system is an environment that is constantly changing and introducing new parameters, even from outside the field. You can't write it down into a table or a check-list. This pandemic as a good example and a representation for the medical field as a whole. From the end of 2019 till today so many things changed: the symptoms, the treatments, tests, protocols, the disease itself. It mattered not only how you felt but also where and when you were and even the geopolitics of the time. The best doctors can keep up and maybe stay ahead of all of these changes and that's what AI doctors will need to do too if they should reach or surpass our level.
It's interesting to think about how an advanced AI of this type would look like. Most systems that became better than humans went through few similar stages as they became better and better:
1. The system is conservative but it can hold its own.
2. It starts creating solutions that people didn't see but can understand their value when they encounter them.
3. It starts creating solutions that people don't really understand but clearly work better than any human solution. This leads to a revolution in the field with people trying to imitate the patterns with varying degrees of success.
4. If it gets really advanced, it can guide people to improve their skill.
5. It can explain what it's doing to a human. This hasn't really been achieved yet but looks like a reasonable goal.
Looking at these I imagine that the first super-human AI doctor will gather information and create treatments that seem more like traditional or tribal medicine, administering off-label drugs and even non-pharmaceuticals without an apparent rime or reason but with good results. I suspect that this will drastically show down its development as soon as it starts moving in this direction because of the strict safety protocols and science based approach in medicine.
Excellent video!
It's amazing that u shared this.... The doctors don't want any innovation except more money
I suffered from Celiac Disease for 2 years and 25 years of Bipolar IV before the Internet saved me. Docs were worse than useless.
Oh no 😞I hope you are doing well today!
Glad you're feeling better. Given that I have been in the situation where one doctor gives bad advice "it'll go away" or such when some medication is needed, the question I have is how many of these doctors know about these diseases?
Your model assumes that they are away of different diagnoses but are trying to pick between them. I would probably guess that a number of the doctors you saw hadn't had a case of the disease that you had. In particular, they still might not know because I do not know if a system that updates the information. I am thinking of supervised learned because if the doctors' hard cases are usually pushed to someone else then they might not ever learn the answers.
For instance the doctor of someone I know was trying to figure out what was wrong with them with some issue and it turned out that someone they knew who worked for a soecialist actually had the correct recommendation: drink water. They were dehydrated chronically. This is the situation you could potentially address with a computer. However, clients not knowing what symptoms they have, honestly tellijg them, or accurately remembering them migt not help... however a computer couod in theory get good at guessing the rigt answer anyway.
Also, I think the doctor should try to make am independent analysis of the situation, then look at the computer. The computer should also have a setting or should generally give a number of possibilities for the illness. This helps upon updatig information and research by the client but also so that the doctor doesn't treat the computer as a calculator and assume that it is giving the correct answer because that is what computers do.
Super interesting idea! So essentially the AI can be used to train doctors? That's a great idea for tightening the feedback loop!
Even when I've approached learning maths from a physics perspective, it still amazes me how many situations you can apply this stuff to.
Right? “Unreasonable effectiveness” and all that
oh I've been diagnosed with endometriosis too, it was dumb of me to not research about it - so thank you for this video. I had gone to the gynecologist after unusual bleeding when I was not on my periods and thought that it was just a one-time disease for some reason. I have extremely painful periods and most of the other symptoms. People aren't kind, even my own mother who knows about it, doesn't really believe me and tells me I'm lazy (when I physically can't move because of the pain I am in). But I don't think I'll have a surgery right now because I'm just 17, I just try to cope by eating pain killers.
But glad to hear your story, I'm so happy for you
I’m sorry to hear about your condition and how the people around you have treated you. It can be so hard to talk about pain, especially when other people can’t see or don’t understand the reason for it. Even then it’s so easy to minimise pain. Sometimes I think about the extreme pain I was feeling at the start of the year and question, “was it really that bad?”, now that I’ve forgotten what was like. But no, the pain from endo really can be that bad, as you know. Try ignore when people are telling you otherwise. They just don’t know.
Unfortunately yes, endo is an ongoing condition, and one you’ll need to keep an eye on :( for now since you’re young you probably can’t/ shouldn’t do much but when it becomes (financially) feasible in a couple of years, I think you should see a doctor specialising in endo, as opposed to a regular gyno. It’s a very complicated disease so you want to be looked after by an expert. Meanwhile though, there’s a lot you can do to help with pain! There’s pelvic floor yoga on RUclips that can apparently help, there’s heat pads, and there’s anti inflammatory period pain medication. I’d really recommend this book to learn more about this condition and strategies to manage it: How to Endo.
It’s not a fun condition but I hope it gets better for you ❤️
@@LookingGlassUniverse @Looking Glass Universe Thank you so much for your tips and concern. Yes, I'll surely read the book to get a better understanding and try out the suggestions you gave!
Ok, good luck :) I hope you feel better soon!
I’ve never been on the backend, analyzing the data that comes in from patients, but I imagine a lot of the information that could be fed into AI systems like these are already collected in psychiatry: it’s pretty normal to be asked to fill out some basic forms reporting your experience of various psychiatric symptoms (persistent feelings of hopelessness, oversleeping, manic symptoms, excessive worrying, etc) that individually can be innocuous (some people fidget a lot, doesn’t necessarily mean you have an anxiety disorder) but together can paint a more detailed picture. If a patient reports long stretches of oversleeping, fatigue, a lack of motivation, and overeating, these depressive symptoms could indicate something like a depressive disorder or a bipolar disorder and I imagine an AI would be well equipped to crunch the numbers from these data and help point the psychiatrist in the right direction.
My brother is a psychiatrist and had the same thought!
Thanks for mentioning possible dependency like C and potential correlation between A and B, because you had previously used P(AandB)=P(A)P(B), but that's only the case if A and B are independent. Hope you are cured!
Absolutely right! This is exactly the sort of dependency causal Bayesian networks are great at dealing with.
Well, I am glad you're doing better and that somebody (or something) figured out what the problem was. What if doctors had to crunch the numbers like that before making a diagnosis? Would that improve their success or have the opposite effect? On the one hand, you'd think it might improve it since they're being more precise. But on the other hand, it might have the opposite effect because there is so much opportunity to make a mistake. Bayesian reasoning isn't easy. I see really smart people accusing each other of doing it wrong all the time. A hybrid solution might be better because a doctor could translate what a person says so that they can better enter the information into the computer, allowing the computer to work better.
Since I'm one of those people who can't read an article on a disease without thinking I've got it, I try not to.
Great point! Doctors being the ones to enter in the patient's symptoms into a bayesian network might produce the most consistent and accurate results.
How are you doing now?God bless you with good health..Macuna
Bless you!
I'm sorry you had to deal with that. From what I hear, endometriosis is very painful and very underdiagnosed.
Glad you got the correct diagnosis and treatment in time and are doing great. I agree that AI has a positive role to play but I would hesitate to suggest that AI alone should be trusted for diagnosis and/or treatment in general. I’m speaking as an engineer and scientist, not just personal opinion. AI is a promising tool and works well but also makes occasional errors. And the unintended or unknown biases in the data that AI is trained on can be crucial, as your calculations have shown. And aside from that, there can be ethical considerations.
Thanks for a great, informative and thought provoking video. So glad to hear from you again! All the best from North Carolina.
I did research in AI for 40 years…
One of the early successes (1970s) was a hand-crafted program, MYCIN, that “knew” about most of internal medicine, problems, and treatments.
It used ad hoc “confidences” rather than proper Bayesian statistics. But it had expert-level performance.
I always thought that AI could provide good assistance to doctors; it would remember rare conditions as well as common ones, for example.
That’s so cool you worked in AI for so long! Why do you think systems (like MYCIN and modern incarnations) haven’t been adopted more widely?
Hope you are doing better. 😢
Can you please make a video on learning physics from books as well as the RUclips playlists that explain deeply ?
Really cool that probability stuff I learnt is use ful
Haha, I know what you mean!
this is so surreal. My husband studied biomedical science and partly due to that, we've just known about endo and assumed it to be a commonly known condition. I'm really surprised that doctors still don't think about it in relation to these kinds of symptoms. I'd hope the younger graduates these days come out of uni knowing more about it. Perhaps it's different here in Australia, but oh well. All the best
actually, just had an idea. Perhaps they DO think about this, but instead they say "not sure what this is, but it isn't normal, let's run some extra tests".
That way maybe they feel they can avoid priming the patient with expectations that may end up being not the case with more testing.
Yeah, I wonder what the thought process was, given these symptoms are very typical for endo and it’s such a common disease. I moved around Australia and the UK so I had a lot of doctors. Some of them even prescribed me medication to help with the pain but weirdly didn’t order any further tests or mention any sort of potential cause. It was only when I got intense pain in my ovary and it wouldn’t go away well past the period that I got referred for an ultrasound. A friend of mine (also in Australia) also has very bad period pain and her doc had prescribed pretty hardcore medication for the pain. After I found out I have endo I asked her to ask another doctor about it. It turned out she did too, but somehow her doctor hadn’t thought of that? It’s really strange that this seems to be a problem with this condition
Hi Mithuna, please make video on job opportunities after becoming quantum physicist??
Your little fan 😊😊🤗🤗
good video, thanks
Glad you're better now!
I think a major problem in medicine is a standardized way to collect data and making it available. Basically, every country and in a lot of cases every municipality or similar granularity has their own system. This makes it very hard to actually gather good data that's statistically significant for anything but the most common diseases. In general doctors readily dismiss many potential causes due to the improbability of them being the case, since "only 1/10000 has it", but the fact is that rare diseases are very common in total, it's just that any given one is rare, and they are also even harder to diagnose due to the rarity itself since there's not as much to go by for identifying it.
And yes, a hybrid approach would be extremely beneficial! The book "Noise" by Kahneman, Sibony, Sunstein, actually talks about this and suggests more hybrid approaches in general whenever human prediction is necessary, because human prediction is _very_ noisy even when the _relevant_ input is the same.
Collecting this data seems like a massive challenge indeed! My brother is a doctor and he was thinking about this. He said that most of the data in his hospital is actually handwritten... not ideal for processing.
Glad to hear you're okay.
For the intuition about P(B|tired) being more likely even though the numbers are rigged up such that they should be equally likely, my guess is that the P(A)=1/4 and P(B)=1/2 somewhat dominate over the P(tired|AB) = 1, P(tired|notA onlyB) = 1/4, P(tired|onlyA notB) = 1/2. The reason I say "dominate" is because the sample size for P(A) and P(B) can be made arbitrarily large while when you're applying the Punnett square probabilities you're within the sample size set the by the initial P(A) and P(B) probabilities.
For example, you mention the sample space of 8 people, with the ratio of 1:1:3:3 for the cases AB:AnotB:notAB:notAnotB. You already see that the notAB sample size is much larger than the AnotB sample size. Now increase the overall sample size from 8 people to 256 people and the difference becomes larger. Contrary to this, while applying the Punnett square probabilities to these numbers, those probabilities are forced to work within the overall sample size (of either 8 or 256). The point I'm trying to make is that making the sample size arbitrarily large helps magnify the differences caused by the initial P(A),P(B), probabilities.
Just my 2 cents. Not sure if this makes any sense lol. Statistics is hard :(
Ah, I love this way of reasoning about it! Yes, you're right, the probabilities here are being capped by the "base rate". That's suc a nice way of thinking about it! It's like in the usual version of Bayes Rule where if p(A)
@@LookingGlassUniverse Ah, cool, glad this makes sense! I'll say though that pondering over it a bit more, I'm not a 100% sure that my explanation is correct. Logically it might make sense, but not sure about its accuracy. So would suggest anyone reading the above explanation to consider it with some skepticism (as with all things in statistics).
Also, it would be great to see the same explanation worked out with different numbers/probabilities. Following your lead, I'll leave that as an exercise for the reader. XD
Endometriosis is a horrible thing to go through. I hope that you're okay.
AI has very similar limitations to diagnostic imaging. It can be a massive problem when the physicians use the output without knowing the limitations of the system, or models used to both encode and then interpret the data. With diagnostic imaging not only does the hardware have limitations such as a unique noise pattern, resolution limit, etc., but when you encode an image you use particular processes and you are /choosing/ to let go of some information, or introduce information that isn't there (averaging for example). This a big deal at tiny, cellular scales and can lead to misdiagnosis, poorer treatment plans and thus health outcomes, radiation dosimetry and the 3D modelling involved is a great example. AI is the same, my professors would constantly tell everyone that it's a black box that not only make people "lazy" or words to that effect, worse still the people using the systems have no real understanding of how the AI came to the conclusions it did.
Intense difficult inspiration for a scicomm outreach special wow thanks… definitely over reliance on various humans or an AI/ML made by I guess humans hahaha is dangerous - worse is perhaps ignoring pain or similar and I’m glad you didn’t do that for too too long - excellent that this AI was so good wow - it does seem that yeah people are willing to be asked a few questions by their phone etc and this is a huge huge help to medical reality etc wow thanks for raising awareness indeed! I definitely, definitely am not expecting to design this AI/ML whatever thing but excellent homework for those who might in particular so thanks for that indeed!🎬🎬🎬🎬🎬🗽☮️💟♾🤯👩⚕️
Glad you're OK! Those photos were scary! :)
I wonder whether "Do you have symptom A?" should itself be a probability rather than a binary value, ranging from "I don't think I have that symptom" to "I almost surely have that symptom." A slider could be presented allowing the user to specify the degree to which they think they might have that symptom, and a doctor reviewing the data could refine those values through discussion with the patient.
I love that suggestion! I wonder how you'd implement that mathematics... do you have any suggestions?
@@LookingGlassUniverse I'm far from being an expert, but naively I would guess that out of each question about symptoms in your tree would come two edges, one for Yes and one for No, with a probability assigned to each (and summing to 1). Then all of the downstream effects of that symptom being present or absent would be weighted by those probabilities. Does that sound vaguely reasonable? :)
the akinator is a great example of this in action
Glad you are OK. Take care 🙂Regards Dr Sabapathy.