Use Case Series: 5. Healthcare's Advanced Analytics & AI Opportunities

The healthcare industry is facing a significant issue with data: there is no interest in sharing it! Or at least, there is a substantial incentive for the public sector, but not so much for the patient. You add insufficient data governance policies and an absence of advanced analytics strategy, and you have a pretty good picture of the state of the healthcare industry.

Now imagine that something insightful is generated for the patient, it would potentially become a strong incentive for a change of behaviour. For instance, a non-invasive continuous glucose monitoring device would lead to an auto-regulated lifestyle for a patient with diabetes.

Shall we use data to control healthcare's cost? Transparent pricing would influence change in the entire ecosystem (e.g. different prices for the same procedure). Or do we believe that consumers are not mature enough to bring healthcare cost down?

AI in healthcare involves the use of computer systems to carry out normal medical practices with little or no human involvement. It has a profound impact -positive and negative- in the healthcare and biotech industries. This impact, discussed below, is largely determined by the type and role of AI technology used. To get a better sense of how AI and machine learning are transforming the healthcare industry today, it’s useful to consider specific cases. AI is transforming the sector in four main areas.

1.  Disease identification

AI aids in faster diagnoses of many diseases. Identifying and addressing diseases and other medical conditions quickly can significantly improve outcomes for patients. One AI technique used in research is imaging analytics. Identifying patterns in images using deep learning technology is one of the strongest points of AI. For instance, in Singapore, Dr Dinesh Visva Gunasekeran is training an AI to diagnose tuberculosis using the ocular features. This approach could bring effective screening and evaluation to TB-prevalent regions that lack radiologists. AI can supplement the skills of specialists by identifying subtler changes in imaging scans more quickly, potentially leading to earlier and more accurate diagnoses.

There are other examples where AI is used to diagnose cancerous cells. Start-up Enlitic is employing deep learning to detect lung cancer nodules in computed tomography (CT) images. Stanford University has developed an AI capable of detecting potential skin cancers from images. AI can efficiently diagnose heart conditions within fifteen seconds. A standard diagnosis performed by a highly-qualified medical expert requires 30 to 60 minutes (Forrest Brown, 2017). AI diagnosis also relies on machine learning, which implies that each use will lead to improved performance. The technology is cloud-based and thus accessible to physicians and experts globally.

It takes an AI less than one minute to diagnose Alzheimer’s disease with an accuracy of roughly 80% based on speech patterns. The AI is trained on features such as the length of pauses between words, the preference for pronouns over proper nouns, overly simplistic descriptions, and variations in speech frequency and amplitude. These factors -imperceptible to human ears- are simple repetitive tasks for a trained AI system.

Google’s DeepMind Health is working with the University College London Hospital to develop machine learning algorithms capable of detecting differences in healthy and cancerous tissues. It helps to improve radiotherapy treatments. At Purdue University in Indiana, researchers are using machine learning algorithms to predict relapse rates for acute myelogenous leukaemia. To measure how promising AI is in the healthcare industry, one of their algorithms was able to identify patients who would relapse with 100% accuracy.

In a world premiere, Singapore National Eye Centre (SNEC), Singapore Eye Research Institute (SERI) and National University of Singapore (NUS) School of Computing have developed an AI to automate eye screening and detect diabetic retinopathy condition in a simple image capture. A small portion of radiology scans can also be automated using an AI, mainly to interpret Chest X-ray.

2.  Drug research

Machine learning also identifies target proteins, which can cause inflammation or lead to tumour growth. Drug research is aimed at designing molecules called ligands, which interact with target proteins to eradicate a disease. Identifying appropriate ligands with current technology is a time-consuming process. Researchers, however, have developed an artificial intelligence algorithm to assist. This algorithm makes use of deep convolutional neural networks for structure-based rational drug design (Wale, 2010).

"Insilico Medicine is finding new drugs and treatments with deep learning algorithms, including new immunotherapies. AI makes these cures work because it can design combination therapies and identify incredibly complex biomarkers by performing millions of experiments in simulated form at lightning speed." (https://sigmoidal.io/artificial-intelligence-and-machine-learning-for-healthcare/)

Through this approach, the AI algorithm (AtomNet) understands complicated concepts by breaking them into small pieces of information. AtomNet has managed to successfully learn the basic concepts of organic chemistry in BioTech simply by studying available data. When pitted against other drug screening techniques, AtomNet compared favourably, implying that AI technology is the future of Healthcare & Biotech sectors.

3.  Treatment recommendation and optimisation

Artificial intelligence also aids in designing better treatment plans. A special program for oncologists launched by IBM Watson provides physicians with evidence-based options for treatment. This program analyses structured and unstructured data in clinical reports to grasp data context and meaning. The program then combines patients’ attributes with clinical expertise and research data to identify suitable treatment procedures. Treatment optimisation based on individual health data will enable personalised medicine with the hope of better disease assessment & outcome. The idea is to optimise the treatment options using patient medical information and history, wearable IoT devices (notably the Apple Watch & the Fitbit) and micro biosensors devices. Not only it will have a significant impact on the patient health optimisation but it will also reduce the overall healthcare cost.

4.  Operation optimisation

Finally, AI also helps to mine medical records. This is the most common use of AI in healthcare. It performs data management by collecting and normalising data, tracing its lineage and storing it in accessible locations. This method is more accurate than conventional data storage, enabling improved and more efficient health service (Powles & Hodson, 2017). Through its Deepmind Health Project, Google is currently researching AI’s role in mining medical records. 

AI assistants (Similar to Siri or Alexa) would help to save a lot of time using voice-to-text transcription to automate or eliminate tasks like ordering prescriptions or writing patient notes. Using natural language processing (NLP), we can now extract semantic meaning from any documents (e.g. unstructured clinical notes, scan of academic articles, satisfaction surveys, etc.) and process search queries written in plain text to return accurate results. Machine Learning can be used to fine-tune the patient risk stratification capabilities, allowing medical personnel to reach out to high-risk patients more proactively and better coordinate their medical support.

The Medical Sieve algorithm launched by IBM is an ongoing project aimed at deriving a “cognitive assistant” with reasoning and analytical capabilities and vast clinical knowledge. At the moment, the Medical Sieve is authorised to aid in clinical decision making in cardiology and radiology. The cognitive health assistant performs analyses of radiology images and detects problems quickly and reliably. This has been most welcome in healthcare and biotech sectors, as radiologists will only be required for supervision of exceedingly complicated cases.

Another usage of AI is for monitoring chronic diseases (based on data from wearables) as well as pattern recognition e.g. to identify outbreaks of infectious diseases within geographical regions, serving as a new public health early warning system tool (think of what we have for Tsunamis/Earthquakes, but instead for the Haze/ Zika virus/ Dengue outbreaks)

5.  Example of an AI application in healthcare

The avant-garde Health Ministry of an EMEA country developed an advanced research environment containing 20 years of high-quality data, to promote health and increase information transparency. They built a platform that allowed the collection, management and analysis of structured and unstructured health information and allowed an optimal balance between privacy and effective information sharing for the benefits of research. The approach consisted of: 

1. Identifying the different research needs and information on analysis capabilities, validation and publication through a qualitative survey among experts in health research;
2. Conducting pilot researches focused on relevant health challenges to demonstrate the capabilities of the advanced analytic models and tools, and to experience the research process (data acquisition, data cleaning, data analysis and modelling) within the platform;
3. Advising the Ministry of Health on topics such as data quality, information sharing and security to build a business model that can be promoted and accepted by data providers such as hospitals, health plans, government agencies and academia.

Wrapping up

Of the many current capabilities of AI, image analysis is the closest to application in healthcare as it centres on pattern-recognition: 

• Something easily trained by the vast amounts of data that already exist and are labelled in a manner, that make for good training data;
• This is also a playing field that the machine has a distinct advantage over the human due to sheer speed and the time it otherwise takes the human to interpret;
• Additionally, the attention to detail and soft signs that are easy to miss are an added advantage to machines in this application whereby the human can get bored with repetitive activities and tired after overnight duty.

Other applications such as Automated Consultations/Translation/Decision making are much further away for a multitude of reasons: 

• Lack of good training data - while vast amounts of data are available, they are heterogeneous without consistent structure/form. Making matters worse, they are structured towards outcomes/labels designed for facilitating insurance claims, which are vastly different from outcomes/labels with clinical significance (which are required for training decision making algorithms for use in clinical practice/point of care);
• Capabilities of NLP still developing - while able to recognise and respond to patterns in intonation/speech, unable to decipher/understand the content of speech. Also, much of the "interpretation" that is conducted with current capabilities relies on contextual clues at least in part, and run a risk of misinterpretation in healthcare whereby contextual clues between vastly different diseases can be the same (eg both stomach flu and cancer can lead to weight loss in someone with a family history of stomach cancer);
• User acceptability of solutions: Healthcare's users are not all tech-savvy, and digital solutions present new barriers such as lack of familiarity as well as misinterpretation of information without correlation with physical cues such as expression/emotion. 

Although there is much apprehension in the sector, AI and machine learning can provide an influx of productivity & revenue gains as well as new hopes and concrete results for patients. This revolution has started to impact the healthcare industry. The use of AI has resulted in more appropriate and accurate treatment plans based on patients’ individual medical histories.

This article is written in collaboration with Dr Hishamuddin Badaruddin, Assistant Prof. (adjunct) in public health sciences at Penn State University.

Hope you like this article!

All the best,

Emmanuel

Please share your thoughts and views by leaving a comment. If you appreciate the insights of this article, please like it and share it with your network. Many thanks!

Disclaimer: The views reflected in this article are the views of the authors and do not necessarily reflect the views of any company or organisation.

#AI #artificialintelligence #DataAnalytics #MachineLearning #FutureofAI #DeepLearning #Heathcare

References:

• Powles, J., & Hodson, H. (2017). Google DeepMind and healthcare in an age of algorithms. Health and Technology.
• Wale, N. (2010). Machine learning in drug discovery and development. Drug Development Research
• Artificial Intelligence And BioTech | ForrestBrown. (2017, August 1). https://forrestbrown.co.uk/news/when-biotech-and-ai-collide-an-exciting-future-for-healthcare/
• Patient readmission model (https://aws.amazon.com/blogs/big-data/readmission-prediction-through-patient-risk-stratification-using-amazon-machine-learning/)