Niraj K. Jha. Professor, Dept. of Electrical and Computer Engineering, Princeton University
The Internet-of-Things (IoT) era promises hundreds of billions of devices or physical objects connected to the Internet. These objects include sensors, actuators, and processing elements that help us gather data, make intelligent decisions, and optimize processes. IoT is expected to have a potential economic impact of more than $10 trillion per year by 2025, with $2 trillion (its largest fraction) coming from smart healthcare applications. Smart healthcare encompasses applications of various Artificial Intelligence / Machine Learning technologies in healthcare. These applications will be enabled by various technologies:
(i) Neural network-based disease detection using wearable medical sensors present in smartwatches and smartphones that communicate with a health server to enable a physician to keep track of a patient’s health. These sensors include Galvanic skin
response, heart rate, skin temperature, blood oxygen, blood pressure, accelerometer, gyroscope, etc. These neural networks can be embedded inside an App in the smartwatch or smartphone. Just one neural network could detect many diseases using multiple heads, one per disease. This can be done very energy- efficiently, thus not draining the battery of these devices much. They could be personalized by deciding which diseases the user is interested in monitoring and attaching the corresponding heads to the neural network body. After the disease is detected, the same models can be used to maintain well-being by flagging extreme manifestations of the condition.
(ii) Interpretability: Decisions made by the machine learning models need to be interpretable in order to foster confidence in their decision-making. A new type of machine learning model, called differentiable logic networks, can provide this interpretability since they are based on Boolean logic even though they are trained in the same way as neural networks. Interpretability is a very important criterion in Medicine to first ensure no harm comes to the patient based on the predictions made by the machine learning model.
(iii) Personalized medical decision-making: Several therapies or interventions are often available for many diseases. Which one is employed for a particular patient is often decided based on trial and error. With the advent of transformers that can capture spatiotemporal characteristics of data obtained from patient cohorts that have previously undergone various therapies, it is now possible to predict which therapy would work best for a particular patient. This opens up a new era of personalized medical decision-making.
Conclusion: The era of smart healthcare is just beginning. There are many challenges that need to be overcome: training machine learning models in low-data regimes, tackling data that are not from the same probability distribution as the data the model was trained on, continual update of the model as new data come in, developing foundation models for healthcare that can be fine-tuned for downstream applications, etc. Thus, this is a great field of research for young researchers to get into.
