Artificial Intelligence in Healthcare
A Report based on the Keynote Speech delivered
– by –
Dr Anurag Agrawal, Director, Institute of Genomics & Integrative Biology, CSIR
@ the Artificial Intelligence in Radiology 2018 Symposium, November 10, 2018
Organized by Telerad Tech and Image Core Lab
In the recently concluded, AIR Symposium 2018 held in Bengaluru which also witnessed the launch of Telerad Tech’s MammoAssist – the AI Software for Early Stage Breast Cancer Detection, Dr. Anurag Agrawal, Director, Institute of Genomics & Integrative Biology, CSIR, expressed his views on Artificial Intelligence (AI) and its game changing paradigm that will help cover the wide gap between the demand and supply sides in healthcare, specially radiology. The topic of his address was “AI in Healthcare”.
Augmented Intelligence or Intelligence Augmentation
Dr. Agrawal said that AI was not new and that it had been brought to light earlier as well. However, the computers earlier could not really implement the true benefit of AI. Over the years, the computational power and the advent of the digital revolution, has given the healthcare sector enough digital data to make these algorithms meaningful. And the potential for AI to completely upend the way the industry delivers healthcare.
Simply put, AI could be defined as making machines think like humans and machine learning was a specific aspect of artificial intelligence. However, he put it in an interesting perspective on augmented intelligence vis-à-vis intelligence augmentation which was not necessarily through machine learning. Bringing in some interesting comparisons, Dr. Agrawal, highlighted key deliverables like AI could work tirelessly for 24×7, could be available instantly to everyone, help in continuous updation, duplicate and grow without ageing – something that humans would lack when compared to these intelligent systems.
Complexity of data surge
Highlighting the limitations of the human mind, Dr. Agrawal stated that, a human mind could only hold five to seven important pieces of information at a time. If that were true, we would soon be approaching limitations of the same and the root cause of our problems in these big data settings would be complexity. As per Dr. Agrawal, a lot would depend on the way we defined our problems in healthcare and its applications. As we acquire more and more data, the size and complexity would defy ease of any categorization. In fact, they couldn’t be even be summarized in human language, accordingly, couldn’t be described and hence humans couldn’t be expected to provide solutions to these problems. For instance, AI could tell the age, gender, detect diabetes, identify if the person was a smoker, show blood pressure and could also predict cardiovascular risks.
To drive home his point, Dr. Agrawal expressed that, through available technologies one could create new systems of prognosis, detection and stratification that have never been done before. However, the doctors themselves would have to let it go, for the field to fully emerge. He highlighted, that the challenge in our country would be posed by the data required to reach accuracy levels – 15,000 images for 80% accuracy to around 40,000 images to achieve over 90% or accuracy.
Dr. Agrawal opined that by creating a competitive market environment annotated data could be made available for free. This could lead to achieving accuracy by reading fewer images. For instance, in the field of deep echocardiography, in only about 100-200 images, the machine was able to read LVH on echocardiograms with 92% accuracy. Invented in 2016, this is a technique called genetic adversary of networks, which creates synthetic images. He explained that a generator in this case, would generate an image which it thought was an echocardiogram, while another network adversarial would check it at the same time. By the time one would reach the last excess post-exercise oxygen consumption (EPOC) of the process, without looking at the annotated image of Left Ventricular Hypertrophy (LVH), the machine would know what an echo was and what a relevant view would be. The learning would then be transferred and with labeled data, one could do a good classification. He lauded, IIT Kharagpur for doing some exceptional work in intravascular ultrasound in this area.
AI and its benefits
Dr. Agrawal expressed that the field of medicine has always been the last one to see any transformation of technology, while fintech has always been the first. In the field of finance, people were implementing AI even before people in the field of medicine were thinking about it. He stated that in the current scenario, AI was already everywhere – in genomics, the machine could dish out lots of relevant stuff, even in radiology, AI was fast becoming the backbone.
Citing examples and comparing them to humans, Dr. Agrawal, said that, AI could bring about positive changes in the healthcare domain. It could describe, classify, predict things and take a step ahead to prescribe things as well for patients. The machines could tell us what to do in order to make a patient better. However, he mentioned, that the most complex things could not be managed by machines alone and would need human interventions.
- Describing things – when a doctor listened to a patient, the focus of the doctor would only be on the part that s/he specializes in as s/he thinks that would be relevant to her/him. S/he would not really be listening to other areas of problems of the patient. However, a machine had the capability to listen to the entire conversation, filter out information about the patient and create them for other relevant doctors.
- Classification – All possibilities exist in clinical diagnosis for basic classifications to be taken care of. For instance, derma level skin cancer detection.
- Prediction – AI lacks the understanding of the likelihood of patients as they walk-in. Dr. Agrawal expressed that we were talking about things we were aware of and not even considering things that we were not aware of and things we didn’t even know existed. Could AI help us with that? Giving an example, Dr. Agrawal highlighted that imagining a human doctor trying to read an MR with 5000 pictures, was simply impossible.
AI – the future game changer
Expressing hope, Dr. Agrawal said, that in the future all the dirty, ultra-complex and background work would be done by machines and AI leaving doctors to do ‘the’ one thing that only they can do and AI cannot – talking to the patient, helping them heal through conversations and informing them properly. He emphasized this human interaction and its importance.
Citing some trends in the field of medicine, Dr. Agrawal said, that AI would impact healthcare services and its quality positively, unlike in the field of biology, as the spirit of sharing existed, and many things would be freely distributed.
He cautioned that regulatory issues would be critical in the future, as we would not want them to become bottlenecks and we would also not want people to be unprotected. As per him, acceptability by consumers and the liability in case of error were a few critical things that required discussions and clarity.
Highlighting the physical patient ratio in our country, Dr. Agrawal hoped that, AI would minimize the impact of low physical patient ratio and prove to be a big game changer, as it would help shift from hospitals to home as a point of contact and reduce the rush at the hospitals. This would essentially lead to shifting from episodic encounters to bringing about continuous surveillance.
In his closing remarks, Dr. Agrawal said that, AI should not be posed as a threat and that it would be positive in the field of medicine bringing out changes required to improve healthcare. However, he warned that, people should not take it lightly, as AI had the potential to replace people who were not communicating properly, were simply in a supervisory role and who would simply make a classification judgment, as AI would make them irrelevant.