Artificial Intelligence and the Future of Healthcare
Priyanka Banerjee20:072021
Identical symptoms can hide different diseases, and data can tell them apart
A boy and a girl walk into a clinic with the same chest pain and palpitations; only an echocardiogram reveals that hers is a genetic heart condition that also endangers her mother. Data scientist Priyanka Banerjee of Speridian Technologies uses stories like this to show what imaging changed in medicine. What happens when a doctor who has never seen the condition gets a machine trained on patients worldwide?
See how machine learning could support diagnosis, readmission risk and mental health care
Data scientist Priyanka Banerjee gives a beginner-friendly tour of how machine learning could reshape healthcare, starting from the history of medical imaging and a story of two patients whose identical symptoms hide very different heart conditions. She shows how models trained on global data could support diagnosis, readmission risk prediction and mental health care, and argues that the main hurdle is getting enough patient data into these systems.
“What is important over here in the medical field is the right treatment during right time.”
“So there are these are the signs which actually what our naked eyes cannot determine, a machine can determine because it has more data inside it.”
What you will learn
Explain why similar symptoms can point to different diseases and how medical imaging helps doctors tell them apart.
Describe in plain terms how machine learning learns from labelled medical images to recognise a condition.
Identify healthcare uses for machine learning, from diagnostic assistance and symptom tracking to readmission risk prediction.
Recognise how natural language processing of word choice in therapy sessions could support mental health care.
Name the main hurdle to AI in healthcare: getting vast amounts of patient data into models so they become more accurate.
Key takeaways from the speaker
Basics of Machine Learning
Problems in healthcare sector that can be solved using Machine Learning
Hurdles require to overcome to implement AI in Healthcare
New technologies that are already using AI in treating different medical conditions
Details
Level
Beginner
Format
Talk
Technical depth
Introductory technical
For
Student, Early career
Length
20:07
Published
June 2021
Questions this talk answers
Why is medical imaging so important for diagnosis?
Before imaging, diagnosis relied on symptoms alone, but imaging showed that similar symptoms can come from different diseases. It is also a main source of medical research today.
How could machine learning help a doctor who has never seen a rare condition?
A model trained on images and patient data from many countries can recognise patterns, such as hypertrophic cardiomyopathy, and act as an assistant that tells the doctor the condition is not uncommon worldwide.
How can machine learning predict whether a patient will be readmitted after surgery?
It compares a patient's features, such as other conditions, vital signs and previous hospitalisations, with historical patient data to estimate the probability of readmission.
How can AI support mental health care?
It can evaluate whether therapy sessions are productive and send feedback to clinicians, and natural language processing can weigh the words a patient uses, such as tired or sleep, to point to underlying issues.
What is holding back AI in healthcare?
Vast amounts of patient data need to be fed to computers so models become accurate, and a dedicated community is needed to accelerate that work.
About this talk
This talk is about potential of Artificial Intelligence in the healthcare sector. AI will increase the ability from mobile coaching solutions to drug discovery, all falling under one umbrella - Machine Learning. AI is already being used to detect early stages of cancer or monitor early stage life threatening episodes of heart rhythms to robotic surgeries and many more. At times VR with AI can create simulated realities to distract patients from pain. It even helps doctors to take appropriate and timely decisions, prioritize tasks, recommend medicines, dosage error reduction and support clinical decision-making with pattern recognitions of health complications. AI nurse can be a chatbot helping patients to cope with life threatening disorders. Robots will not only help during surgeries but also for repetitive tasks and to a great extent in rehabilitation centers to support people with long term disorders.
With all these possibilities and developments, health care domain will get revolutionized and might become "one giant leap for mankind".
A diagnostic assistant trained on the world's data (86 s)
Words as signals in mental health care (51 s)
Hello everyone. I hope you all are having a great time attending this global conference. So today I'm going to talk about artificial intelligence and future of healthcare. I'm Priyanka. I'm basically from Bangalore. So I have been born and brought up in Calcutta. So while I was born and brought up, I grew up over there. I got to realize there are certain stories and hidden events which we barely know about Calcutta and those amazing stories I got to hear from my grandparents. While I was hearing those stories, I realized the fact that they have instilled a particular kind of education, like education is very important for solving any problem in their children. And finally, that has been passed to us, our grandchildren. So they have always
told us education is the answer to solve the greatest problems existing in the world. So during the early 21st century, when I was passing out of my college, of my school at that time, it was a time when I had to consider which subject shall I choose. So at that point of time, it has been considered engineering or doctor or the two professions which can make a person successful in life. And as a result, I had taken up the technical field.
And there are several friends of mine who had actually gone ahead and taken up the medical field. So while this 10 years down the line, as different technical parts have unfolded, and I have come across how technology has advanced with time, I also got to realize, okay, with the technical advancement, our life is becoming simplified with time. So nowadays, technology has advanced to such a level.
Today, we are standing in such a position where we can just ask Alexa to book a hotel, to book a flight, to place a reminder for our medicine, like, it acts like just like our assistant. So it has actually simplified our life to a great extent. Now, a couple of years back, when I met with one of my friends, who is a doctor, and while we were having a conversation over a cup of coffee, I got to realize that technology has actually advanced a lot and simplified our normal day-to-day life.
But for the doctors, it didn't. Their tasks, technology didn't put much impact over there. They have to go through the critical structure flow of their daily routine. And they actually require it. That is when I came up with this realization, like technology and artificial intelligence, which is a booming technology nowadays, has to play and has to come in the healthcare domain.
So while I'm in this sector of healthcare domain, let us first analyze what is health. Health is nothing. If I just do a Google search, I will get to see it is the state of being free from illness or injury. It basically states what is a person's mental and physical condition. So it is very important to understand health. By health, we just don't mean the physical health.
It is also the mental health because as the saying goes, a healthy mind dwells in a healthy body and vice versa. Definitely. Now, if I show you this picture to everyone of you, what is this picture? What is the importance of this picture? I do not know how many of you know, but this picture actually brought one of the revolutionary change in the field of medicine.
Why? Because this picture is the first ever picture taken, which has given us a peak inside human body. Before that, we didn't know how a human body used to look like from inside. So this is the first picture of an X-ray. And the person who had taken this picture, Sir Wilhelm Röntgen, he has been awarded with Nobel Prize in the year 1895.
So you can understand how important this picture is and how important part this picture has played in the history of medical science. Now, why imaging is so important? Imaging is so important because let us see a scenario before imaging. Before imaging came into existence, people generally used to do any kind of diagnosis based on the symptoms. Now, I can visit a doctor and I can tell, OK, I'm having some pain in my hand and other person can also tell the same symptoms.
The doctor, based on his experience over time, will prescribe the medicine. But imaging actually gave us the first in look inside a human body by which we got to know, OK, similar symptoms may not have the similar kind of diseases. Diseases might vary at that point of time. As a result, imaging plays a very important part in determining different kinds of diseases for similar kind of symptoms.
And also along with that, it is the main source of research nowadays. So people, if this imaging wouldn't have happened, people wouldn't have known like, OK, for similar symptoms, different types of diseases existed. Now, the name of this particular types of imaging, which is present over here. I believe like one or the other we are acquainted with, especially X-ray.
I still remember during my childhood when I was riding my first bike, I had fallen down and I had to visit a doctor. I had a pain in my hand and he had asked me to take the first X-ray of my hand. And that is when I got acquainted with what is X-ray. So this is actually, I want to show you a scenario, how actually imaging has helped us and how machine learning going ahead can help us.
So imagine a scenario, a boy and a girl visits a doctor. Now, both of them complain of similar kind of problems that they are experiencing, which is palpitation, chest pain and difficulty in breathing. After analyzing their problems, the doctor won't go ahead and prescribe medicines. First, her work will be to check the vitals of the patients. He will check the vitals.
Let us consider the boy is having different vitals, like the vitals might be different from the normal ranges, but the girl is showing similar vitals of normal people. Now, the doctor is a bit confused. What is wrong with the girl? Even after showing the similar vital, she is having these problems. She will go ahead and ask the girl to do ECG and echocardiogram.
So when the eco reports came for both the patients, the doctor can analyze and see, okay, for the boy, it is a hypertension. He is having a problem. Maybe due to his lifestyle or the way he work or the stress that he takes in his work. That is where going ahead, he has got hypertension. But for the girl, there is altogether a different problem.
The girl is having a condition called hypertrophic cardiomyopathy or HCM. What is this? This is basically the division of the wall that divides the left and right side of her heart. That is a little bit thicker in her case than normal people. All of us have normal ranges of each and everything in our body. But in her case, the muscular wall that divides the left and right side of the heart, that is more thicker, thickened in her case in the ventricular sector.
So when the doctor is seeing this, he can understand, okay, this is a problem she has got. And this is generally genetic in nature. This doesn't have to do anything with your lifestyle. It doesn't have anything to do with the bad habits which we are having. So for all this while, she had this problem from her birth.
But she got diagnosed with this problem when she is around 20 years old, when she suddenly had started having these symptoms. Now when the doctor got to know she is having this problem, he will accordingly provide her with different medications. But for the boy, the medication will be different for hypertension. So we can see similar symptoms having different set of problems over here.
That is where machine learning also plays an important role. How? Since now the doctor got to understand she is having a genetic disorder, she can ask her parents to get the same checkup done. If her parents will do the same checkup, there can be a probability that one of her parents is also having this disorder from where she has received it.
And after doing that checkup, they got to see, okay, her mother is also having the similar disorder. So now her mother, now people with this kind of problem, they are always at a higher risk of experiencing an irregular heartbeat, which can be life threatening. Now when, now her mother is at around 50 or 55 years old, when doctor got to realize her mother is also having similar kind of problem, but it didn't show any symptom.
Since her age is around 55 years, she is at a higher risk of experiencing an arrhythmia or irregular heartbeat, which can be light threatening. So to prevent that, the doctor will place an ICD or it's an instrument like pacemaker, which will be placed over here on the shoulder, which will control her from experiencing a life threatening heartbeat.
Now why is it required? A life threatening heartbeat is something within a fraction of second, it can take away your life, but that ICD will actually help you. Okay. If you are experiencing certain kinds of heartbeats, it will give you that relaxation. It can restart your heart within that exact time. What is important over here in the medical field is the right treatment during right time.
So ICD, when it is placed over here, it will do the right treatment during the right time. It will, but life threatening heartbeats are so risky. It will not even give you that time to take the patient from your house to the nearby hospital. So this way it saved us. Now we can understand this image of the echocardiogram of the heart actually saved not only the daughter's life, but also the parents life.
So that is why we told it is a revolutionary change which has been brought by imaging in the medical field. Now, what is machine learning and how that is going to impact from this image? Machine learning is basically teaching the machine to learn. So when you are teaching the machine to learn about certain data, like I will feed this kind of data and I will tell the machine, okay, if you get to see this kind of images, you can determine at that point of time that basically this person is having HCM.
Now, let us take an example that a boy in Africa is having this problem. Now he is visiting a doctor, the doctor whom he has visited, the doctor didn't get to observe or get to know this kind of patients before. As a result, it became unlikely for him to diagnose what is the exact problem. Now, imagine in case of the doctor, if he had an assistant as a machine or as an application.
That machine actually has data globally from my country, India, from USA, from London, UK, from Germany, from different parts of the world. So it can easily determine at that point of time, okay, the problem which the boy is having, that is not something uncommon. Many people around the globe had this problem and the symptoms and the image that has been shown up, that determines that the boy is having HCM or hypertrophic cardiomyopathy.
So we can see how machine learning is actually making the job of the doctor easy over here. So symptoms based treatment followed by medical images. When medical images are run through different machines, it will definitely revolutionize the health sector. In this way, image processing is going to play a very vital role in the part of medicine. Now the question comes how it is going to be helpful for the doctors.
One way we have seen how it can become an assistant for the doctor, determining like what is the problem from the images. Also, there is another way, like if I built an application where I can keep a track about I am experiencing certain pain in my hand. From when am I experiencing that pain if I can keep that track?
What am I feeling when that exact pain I am having? How was my vital? Is any other thing I am having along with that pain? If I can keep a track of all these records on an application and I can provide that to a doctor when I am visiting him after four or five days, it will become so smooth and so easy for him to determine what is my exact problem.
Another place where machine learning can play an important role. Also, along with that, there are certain cases where it can prove to be very, very useful for the doctors. And this is something which Google has done a survey over here and they had published a paper also on the same thing. So the main problem over here is, say, suppose a doctor, a patient has gone through a surgery and it's a very crucial surgery that he has gone through, like a bypass surgery or some transplant or chemotherapy, or there can be several types of surgery or crucial patients getting admitted daily.
Now, what is the risk of that patient getting readmission very soon? How can the doctor determine post-surgery that the patient, maybe within five, six days, he can get sick again? Now, on that patient's previous data of, say, thousands of records they will be having. So on those data, the machine can learn how the patient has behaved over time.
And based on it, it can come up with a probability like, OK, in this cases, it will compare like, OK, this particular patient A has this many features of this many problems based on the data points that he has. Like, say, he is having certain other medicines. For example, a patient had a bypass surgery, but that patient is also having thyroid problem.
Along with that, his vital signs are not normal and he had been hospitalized twice before. So in this way, we can get certain data which he is already having. Based on those data, he will compare with other patients' data that is already trained on the machine. Those data comparisons will give him an idea if those patients, the other patients on whom the data has been trained, if they got readmitted again or not.
If the readmission rate is calm, very less, in that case, the rate over here, like the chances of read admission, the probability will be very, very less. If that is very high, like in case of any transplant patients, in that case, this number will increase. So there are these are the signs which actually what our naked eyes cannot determine, a machine can determine because it has more data inside it.
So as I have spoken about physical health for now, equally mental health is also essential. The first is like the therapist whom we visit when we are having a kind of problem mentally, we are upset. It is very essential to understand what kind of therapist we are visiting. Many times it has happened with me, like I have visited the right therapist, I believed, but after the conversation, I didn't feel right.
I felt like okay, I need to have a session with some other therapist, you know. Again, evaluating a session, if it is fruitful for the patient or not, and sending feedbacks to the clinician. By this we mean, say, suppose I'm having a conversation with a therapist, but it has been more like a chitchat conversation rather than extremely productive conversation.
Like he didn't ask me structured questions from which he can gain certain views about me. So that is also possible. So as a result, this kind of evaluation helps you or will help any clinician or any therapist to understand who the patient or what exactly the patient requires. Because anyway, mental health is such a sphere where people have to work and researchers have to take place more and more.
Even from the choice of words where natural language processing can play a vital role. From the choice of words that we make while in the sessions with the counselors, we can provide parameters. Like if someone is using words like anger or tired or sleep disorder, this kind of words, how many times he is using, what is the weightage of each words.
Based on that, we can understand, okay, he is using more of sleep deprivation. So we can understand, okay, his depression is related to that. He is insomnia. This kind of important role machine learning can play in mental health and help the psychiatrist to use, to make all these sessions with the patients extremely fruitful. These are some articles and use from newspapers that have got an extremely fruitful one.
So now it is Apple Watch has come up with an ECG function where it gets to check if a person's heartbeat is normal or not. So it has actually been proven life saving for several patients. This is a guy whose picture we can see over here. This person had been admitted in the hospital during the right time.
He had actually got arrhythmia or irregular heartbeat, which we call twice. And being a responsible citizen, he visited the cardiologist and they did an ecocardiogram and got to visit okay. He is having a problem for which his valve in the heart is not working or functioning properly. And he needed replacement then and there. It was an emergency surgery.
And he had done that and that proved to be fruitful for him and his life got saved. So in this way, machine learning has proven already started impacting people's lives and saving their lives. With that, I want to see and conclude that these are the certain challenges that we are facing nowadays in making or training the computers.
So we already know computers are there and advanced technology is there in several hospitals globally. But the problem is this vast amount of patient data need to be fit to the computer so that the computer become as smart as humans. And as a result, a dedicated community should be there who can accelerate and bring more benefits in this kind of work so that more patient data can be fit in the computer and the model can become more accurate with time.
So the vision is to make the doctors work much easier over time so that people by sitting at home, they can talk to a doctor along with that professional or a chatbot now it is whom we talk to that kind of help or assistant we can talk. And as a result, the life of the doctors will become much easier.
So that is a mission. And I believe with harnessing the power of data and artificial intelligence that mission can become reality very soon. Thank you.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
How can machine learning help doctors tell similar symptoms apart?
Data scientist Priyanka Banerjee traces a line from the first X-ray to models that could assist diagnosis, predict readmissions and support mental health care.
In short
Priyanka Banerjee, a data scientist at Speridian Technologies, argues that technology has made daily life easier for most people but has barely touched the demanding routines of doctors. Using the history of medical imaging and a story of two patients with identical symptoms, she shows how machine learning trained on large, global patient data could help doctors diagnose earlier, predict risk and support mental health care. The biggest hurdle, she says, is getting enough patient data into these systems.
Key takeaways
Basics of machine learning
Problems in the healthcare sector that can be solved using machine learning
Hurdles that must be overcome to implement AI in healthcare
New technologies that are already using AI in treating different medical conditions
Technology simplified life, but not for doctors
Banerjee opens with her childhood in Calcutta, where her grandparents taught that education is the answer to the world's greatest problems. Choosing between engineering and medicine, she took the technical path while several friends became doctors. Years later, over coffee with one of those friends, she noticed a gap: voice assistants could book flights and set medicine reminders, yet a doctor's working day still followed the same demanding structure.
"That is when I came up with this realization, like technology and artificial intelligence, which is a booming technology nowadays, has to play and has to come in the healthcare domain." (2:35)
She also defines health broadly, reminding the audience that it covers mental as well as physical wellbeing, a point she returns to later in the talk.
Imaging showed that similar symptoms hide different diseases
To explain why data matters in medicine, Banerjee goes back to the first X-ray image, credited to Wilhelm Röntgen, which for the first time let people see inside the human body. Before imaging, she explains, doctors diagnosed from symptoms and experience alone.
"But imaging actually gave us the first in look inside a human body by which we got to know, OK, similar symptoms may not have the similar kind of diseases." (4:56)
She brings this to life with a scenario. A boy and a girl both arrive with palpitations, chest pain and difficulty breathing. The boy's vitals are abnormal, and tests point to hypertension linked to lifestyle and stress. The girl's vitals look normal, so the doctor orders an ECG and echocardiogram, which reveal hypertrophic cardiomyopathy (HCM), a thickening of the muscular wall dividing the heart, which the speaker describes as generally genetic. Testing her parents shows her mother has it too, without symptoms, and at an age where a dangerous irregular heartbeat is a real risk. An implanted device can restart the heart in the seconds that matter.
"What is important over here in the medical field is the right treatment during right time." (10:03)
Teaching machines to read medical images
With the stakes set, Banerjee introduces machine learning in the simplest possible terms: show a model many labelled images and it learns to recognise the pattern of a condition such as HCM.
"Machine learning is basically teaching the machine to learn." (10:56)
Her example is a boy in Africa whose doctor has never seen an HCM patient. A diagnostic assistant trained on data from India, the USA, the UK, Germany and elsewhere could recognise that the boy's symptoms and images match a condition seen many times around the world. In her view, symptom-based treatment followed by images run through such models will make image processing a vital part of medicine.
Predicting readmission and supporting mental health
Banerjee then widens the lens. Patients could log their own symptoms, timing and vital signs in an app, giving doctors a clearer history at the next visit. She also describes readmission prediction, which she says Google has studied and published a paper on: after a serious surgery, a model compares a patient's other conditions, vital signs and previous hospitalisations with thousands of historical records to estimate the chance of returning to hospital soon.
"So there are these are the signs which actually what our naked eyes cannot determine, a machine can determine because it has more data inside it." (15:28)
On mental health, she draws on her own experience of therapy sessions that felt more like chitchat than structured conversation. Models could evaluate whether a session was productive and send feedback to the clinician. Natural language processing could weigh how often a patient uses words such as tired, anger or sleep, pointing a psychiatrist toward issues like insomnia behind depression.
"Even from the choice of words where natural language processing can play a vital role." (17:08)
The data challenge and the vision
As evidence that this is already happening, Banerjee shares a news story about the Apple Watch ECG feature flagging an irregular heartbeat in a man who then had emergency heart valve surgery that saved his life. She closes on the hurdle: hospitals already have computers, but vast amounts of patient data must be fed into models before they become accurate, and a dedicated community is needed to speed that up. Her vision is a future where people can consult a doctor or a chatbot assistant from home, and doctors' work becomes much easier.
"And I believe with harnessing the power of data and artificial intelligence that mission can become reality very soon." (19:51)
Watch the talk
Artificial Intelligence and the Future of Healthcare, Priyanka Banerjee, 20 minutes: https://www.womentech.net/video/artificial-intelligence-and-future-healthcare
About the speaker
Priyanka Banerjee is a Data Scientist at Speridian Technologies, based in Bangalore, with ecommerce domain expertise and a deep interest in modeling large volumes of data to explain complex problems of business significance. She is currently exploring state-of-the-art Hugging Face machine learning models in natural language processing.
Questions this talk answers
Why is medical imaging so important for diagnosis?
How could machine learning help a doctor who has never seen a rare condition?
How can machine learning predict whether a patient will be readmitted after surgery?
How can AI support mental health care?
What is holding back AI in healthcare?
Drafted with AI from the talk transcript. Editor review pending.
Cite this talk
Priyanka Banerjee. "Artificial Intelligence and the Future of Healthcare." WomenTech Network, 2021. https://www.womentech.net/video/artificial-intelligence-and-future-healthcare