AI Revolutionizes Healthcare
Recent advancements in AI have ushered in a new era of healthcare, with breakthroughs in deep learning, computer vision, and natural language processing enabling AI to assist and, at times, even outperform healthcare practitioners. In this article, we explore how AI is transforming clinical pathology, medical diagnostics, therapeutic development, infectious disease surveillance, and patient management, improving outcomes for millions of people.
AI Accelerates COVID-19 Intelligence and Therapeutic Development
In 2019, the White House Office of Science and Technology Policy announced the launch of COVID-19 Open Research Dataset. CORD-19 is the most extensive machine-readable coronavirus literature collection available for data mining. The dataset was created through a coalition of organizations, including Microsoft Research, the National Library of Medicine, Kaggle, the Semantic Scholar project by the Allen Institute for AI, and the Chan Zuckerberg Initiative. The aim of CORD-19 was to encourage cross-institutional AI research and therapeutic development by enlisting the global AI community. The objective is to use natural language processing and deep learning to surface vital medical insights from over 128,000 scholarly articles about COVID-19, SARS-CoV-2, and related coronaviruses. This vast amount of data would be impossible to analyze without the fast parsing and pattern recognition capabilities that AI brings to healthcare research.
Another example of AI-driven efforts to combat the COVID-19 virus was the discovery initiative launched by Google-owned DeepMind. The company has developed AlphaFold, a Deep Learning model which can predict the protein structure of COVID-19. Traditionally, it would take millions of hours to enumerate every potential protein configuration to reach the 3D structure. However, AlphaFold leverages decades of prior research by parsing large genomic datasets to release protein structure predictions of under-studied proteins associated with SARS-CoV-2.
Numerous AI initiatives aimed at discovering an effective treatment for COVID-19 have emerged globally. For examples, Project ‘IDentif.AI’ (Identifying Infectious Disease Combination Therapy with Artificial Intelligence) was launched at the National University of Singapore to rapidly identify drug combinations capable of treating COVID-19 infections. According to Professor Ho, the leader of Project IDentif.AI, it only took three days for IDentif.AI “to identify multiple optimal drug regimens out of billions of possible combinations that reduced the VSV infection to 1.5% with no apparent adverse impact. This speed and accuracy in discovering new drug combination therapies are completely unprecedented.”
AI has playing a critical role in combating COVID-19, from predicting the virus’s protein structure to discovering new drug combination therapies in record time. These initiatives demonstrate the unparalleled speed and accuracy that AI can bring to healthcare research.
AI Helps Triage and Monitor Patients in Hospitals
AI-based systems have proven to be invaluable in clinical settings, saving critical resources and time for healthcare practitioners and patients.
For example, during the COVID-19 pandemic, AI radiology solution company Aidoc was approved by the FDA to detect COVID-19 patients. Aidoc’s model can identify incidental findings associated with COVID-19, such as unique lung abnormalities typically associated with viral pneumonia, and accordingly, triage patients. In China, Alibaba has trained a model capable of accurately diagnosing patients from CT image analysis 60 times faster than human detection. According to the company, the model can assist in identifying characteristics of coronavirus pneumonia in CT scans with about 96% accuracy, and the entire test only takes 3 to 4 seconds. Alibaba has opened anonymized datasets to health practitioners worldwide to enable better prediction of the pandemic’s evolution and more effective patient management.
Hospitals are also increasingly relying on AI-powered CT image analytics to accurately triage COVID-19 patients. The Royal Bolton Hospital (UK) is harnessing AI insights from the global player in AI-driven radiology Qure.ai. The system “automates the interpretation of COVID-19 proliferation from chest X-rays, making it easier for healthcare professionals to monitor the extent and rate of progression of the viral infection.” (source: Bolton NHS Foundation Trust) The Royal College of Radiology published implementation guidelines for the adoption of AI systems and platforms designed to recognize abnormalities in images and trigger alerts for emergencies.
AI is not only being deployed to predict which COVID-19 patients should be transferred to the ICU, but also to monitor patients post-transfer. For example, a six-year collaboration between Stanford University and Intermountain LDS Hospital resulted in Deep Learning models being trained to accurately detect high-risk postures and falls of ICU patients. To respect privacy, depth sensors were used to capture only silhouette data. Algorithms correctly identified mobility activities with an accuracy of 87%. AI-based systems can detect when a patient has fallen or is likely to fall and alert the healthcare staff for immediate assistance. This alleviates nurses’ concerns about leaving a patient unattended while caring for others. (MIT)
AI-Powered Precision Diagnostics Speed up Time to Detection
In addition to assisting in emergency rooms, AI diagnostics have shown to be highly effective in detecting and predicting medical conditions, such as various forms of cancer. A significant report from The Lancet Digital Health evaluated the “performance of Deep Learning models to be the equivalent to that of healthcare professionals”.
In 2019, a joint effort by MIT and The Massachusetts General Hospital resulted in the development of an AI model that predicts the risk of developing breast cancer within five years. The algorithm demonstrated equal precision for both white and black patients, a notable improvement over previous tools. Trained on 90,000 mammograms from diverse patients, the model’s equitable diagnostic accuracy is particularly crucial given that African-American women have a 42% increased risk of dying from breast cancer, often due to systemic barriers to prevention and treatment.
In January 2020, Google Health also demonstrated the superior effectiveness of its algorithmic model in breast cancer diagnosis, surpassing radiologists’ expertise. Researchers from Google Health and Imperial College London teamed up to train a Deep Learning model on a dataset of X-ray images from 29,000 women. According to an independent research study published in Nature, the model outperformed all of the human readers. The AI system not only decreased the workload of the second reader by 88%, but the study found it able to surpass human experts in the prediction of breast cancer. (source: Nature). “Our team is really proud of these research findings,” said Dominic King from Google Health. “We are on our way to developing a tool that can help clinicians spot breast cancer with greater accuracy” King added (source: BBC).
Furthermore, Google presented an Augmented Reality Microscope with Real-time Artificial Intelligence Integration that improves the accuracy of cancer diagnosis. “The platform consists of a modified light microscope that enables real-time image analysis and presentation of the results of machine learning algorithms directly into the field of view” (source: Google). By retrofitting the technology into low-cost light microscopes, this cost-effective and highly accurate approach enables the seamless integration of AI into routine workflows, democratizing Deep Learning adoption for pathologists worldwide.
AI Improves Patient Outcomes
As the oncology drug market continues to grow, with annual spending reaching $200 billion, predicting patient responses to treatments remains a complex challenge with limited evaluation options. To address this issue, researchers are turning to multivariable AI models to predict the sensitivity of tumors to systemic cancer therapies, in an effort to optimize treatment outcomes.“With AI, cancer imaging can move from an inherently subjective tool to a quantitative and objective asset for precision medicine approaches,” said Dr. Laurent Dercle, an associate research scientist in the Department of Radiology at Columbia University Irving Medical Center.
AI prediction models have a significant advantage over human capabilities as they can learn from vast and increasingly diverse patient datasets. This enables them to continuously identify new patterns, which plays a substantial role in improving cancer care.“These systems are able to learn from each incremental case and can be exposed, within minutes, to more cases than a clinician could see in many lifetimes.” (Buch V., Ahmed I., Maruthappu M., 2016).
Lunit, a precision medicine company, exemplifies the positive impact of AI on patient outcomes through its development of AI-powered therapeutic biomarkers. These biomarkers can identify the most suitable treatment and predict patient response with 90% accuracy for breast cancer pathology images. AI’s superior efficacy results from the widespread adoption of digital technology in healthcare, including the digitalization of histology images that enables AI models to identify and quantify cancer tissues and aids in the development of AI biomarkers for cancer therapy.
AI is showing remarkable promise in the field of medical diagnostics, and a notable example is Google DeepMind’s application of Deep Learning to predict Acute Kidney Disease (AKD). By analyzing a large anonymized electronic health record dataset, the model can predict AKD 48 hours ahead of doctors with 90% accuracy. This technology has the potential to protect patients from reaching life-threatening stages of the disease, thereby avoiding invasive procedures like kidney dialysis. According to Pr. of Renal Medicine, Donal O’Donoghue, this is a game-changer that will advance understanding and care of acute kidney injury. Google DeepMind plans to extend the predictive model to other life-threatening infections, including sepsis.
Using AI prediction also impacts mental health patient outcomes. Science 37, the industry leader in virtual clinical trials, and AiCure, a leading AI company, recently announced a collaboration to deliver AI-powered virtual trials. “As the first endeavor for the new partnership, the two companies will use a virtual or decentralized research model — in which patients participate from home — to evaluate an investigational treatment for Major Depressive Disorder (MDD) in a clinical trial this summer.” (source: AiCure) AI systems are utilizing advanced technologies such as Speech Analysis, Facial Analysis, and Deep Learning to gather and analyze datapoints related to patients’ expressiveness, psychomotor function, and cognitive abilities. These systems use this information to predict the presence of a pathology or determine whether a patient is responding to treatment.
AI-Powered Telehealth Democratizes Healthcare Access
The World Health Organization estimates that two-thirds of the world’s population has no access to diagnostic imaging. AI shows potential to address healthcare inequalities by providing diagnostic expertise to primary care in low-income countries with inadequate infrastructure and limited healthcare personnel. Intelligent radiology tools, powered by AI, enable remote screening and make specialized care more accessible to primary care physicians, reducing healthcare disparities. However, to fully close the gap, datasets must become more representative of diverse populations and disease presentations. As this happens, AI tools will democratize healthcare and improve access to care for people worldwide.
Progress is not solely brought by algorithms. In developing nations, smartphone images are becoming a reliable supplement to clinical quality imaging. For instance, researchers at UTHealth diagnosed skin cancers, with accuracy rates up to 90%, using low-cost “smartphone microscopes”.
Relatedly, AI company DeepGestalt relies on smartphone photos and Facial Analysis to diagnose rare syndromic genetic conditions. Through its medical app Face2Gene, ”DeepGestalt achieved 91% top-10 accuracy in identifying the correct syndrome on 502 different images. The model was trained on a dataset of over 17,000 images representing more than 200 syndromes,” (source: Nature).
AI is revolutionizing access to diagnostics and care across various specialized fields. FDA-approved mobile health technology is now bringing AI to primary care offices, enabling the automatic detection of diabetic retinopathy pathologies. Early and accurate detection is crucial for conditions such as retinopathies, making this a cost-effective and critical development for patient outcomes. The IDx-DR, an AI-powered medical device for detecting diabetic retinopathy, has been approved by the FDA as a Breakthrough Device. This designation is given to medical devices that offer significant advantages over existing approved or cleared alternatives.
Low-cost diagnostics assistive AI apps are set to transform access to care. Harvard Medical School uses Buoy’s AI, an intelligent chatbot that listens to patients to diagnose and provide guidance to patients. Babylon by Telus Health delivers fast personalized and interactive care via an AI chatbot Symptom Checker. A comprehensive health record, a knowledge graph, and a probabilistic graphical model are utilized by the underlying system to predict a patient’s possible condition.
Conclusion
As we move into an era of intelligent automation, the deployment of AI into healthcare holds considerable promise for precision medicine and medical research. Machine intelligence, open data, and sophisticated technologies reinvent healthcare through prediction and superhuman analysis. Undoubtedly, the benefits of these advancements will optimize public safety and individual care. As AI technologies open a new era for telehealth and reinforce disease monitoring strategies, we must ensure that privacy, civil rights, and civil liberties are protected. The Hippocrates’ oath, by which generations of healthcare practitioners have been sworn in, includes a privacy seal that the data revolution shall not supersede: ”I will respect the privacy of my patients, for their problems are not disclosed to me that the world may know” (Hippocrates, 460–370 BC). Medical ethics must transfer to AI-powered healthcare, placing it under the governance of the timeless principle, Primum Non Nocere — ”First do no harm.”
Originally published by Dr. Valerie Morignat at http://intelligentstory.com.
