Computer simulation of human behavioral intelligence once sounded like the stuff of science fiction thrillers. Today, artificial intelligence or AI has made its way into our daily lives in the form of internet search engines, social media, mapping apps, and smartphone facial recognition. It is gradually becoming integral to healthcare, as well. By 2023, the AI medical imaging market1 is projected to reach $2 billion. Currently, about 17% of providers use AI technologies2, and another 30% plan to do so in the future. Though still in varying stages of development and use, AI in the healthcare industry is gaining momentum due to the vast potential for machine learning to enhance diagnostic processes, as described in these examples.
Recently, Brian Wallach, a former Obama staffer was diagnosed with amyotrophic lateral sclerosis (ALS) at age 37. Nearly 6,000 Americans receive this diagnosis3 each year.
Survival rates among patients with progressive neuromuscular disease vary dramatically, and predicting survival based on clinical characteristics is challenging. A study cited in Neuroimage Clinical4 determined that deep learning is beneficial in disease prognostication.
In the study, survival categories were predicted correctly in 68.8% of cases that used deep learning based on clinical parameters. Deep learning based on magnetic resonance imaging (MRI) predicted 62.5% correctly using structural connectivity, and 62.5% using brain morphology data. Based on the study, when researchers combined all three data sources, deep learning prediction accuracy increased to 84.4%.
The progressive degeneration of the brain known as Alzheimer’s disease is the most common form of dementia5. While Alzheimer’s is prevalent in those over age 65, about 200,000 Americans are diagnosed with the early-onset form of the disease.
Researchers have used 18-F-fluorodeoxyglucose positron emission tomography (FDG-PET) technology to train a deep learning algorithm to predict Alzheimer’s6, years before a conventional diagnosis can be made. The FDG radioactive compound is injected into the blood, and its uptake into brain cells, a gauge of metabolic activity, is measured in the PET scan.
Once symptoms have manifested, loss in brain volume is significant. However, with earlier detection through pattern recognition association with accumulation of beta-amyloid and tau proteins, there is potential to slow or halt the progress of the disease.
3) Musculoskeletal injuries
The aid of AI could help radiologists identify fractures7 and other bone issues more readily. The goal of prompt, confident treatment is to shorten hospital stays and hasten the patient’s return to mobility. It also reduces the radiologist’s legal risk.
The unbiased algorithms of AI radiology tools detect subtle variations in images that may indicate:
- Hidden fractures (potentially overlooked in trauma-related emergency department visits)
- Soft tissue injuries
- Loosened joint replacement devices (rather than relying on multiple X-rays over a span of years to indicate a progressive problem)
A study published in the Journal of the National Cancer Institute8 showed that AI technologies perform at levels which rival radiologists in the evaluation of digital mammography. Authors also found potential value in AI as an answer to a shortage of radiologists in some countries.
In addition to breast cancer evaluation, AI will likely make strides in the future of cancer care through:
- Automation of processes in the initial interpretation of images
- Shift in clinical workflows
- Decisions on intervention
- Assessment of the impact of disease and treatment on adjacent organs
- Prediction of clinical outcome
- Continued observation
5) Cardiovascular problems
Automation can help to reveal heart abnormalities in imaging tests such as chest x-rays. An individual’s risk for cardiovascular disease is determined by measuring different structures of the heart. A chest x-ray is often the first imaging study performed when a person complains of shortness of breath or other symptoms related to heart disease.
According to the American College of Radiology Data Science Institute9, AI could be taught to make an accurate assessment of cardiomegaly enlarged heart by viewing just one anteroposterior or posterior-anterior chest x-ray. That allows radiologists to rule out other pulmonary and cardiac problems and reduces the chance of inaccurate assessment of cardiomegaly.
6) Thoracic conditions
Radiology images are often used to diagnose thoracic disorders such as pneumonia, and to rule out other lung conditions. Pneumonia can be difficult to identify, especially in a patient with pre-existing lung issues.
A recent study in the JAMA Network Open10 showed the value of a deep learning algorithm in the identification of chest x-rays with the thoracic disease. The deep learning-based automated detection (DLAD) algorithm consistently kept pace with medical professionals in determining which chest radiographs showed major thoracic diseases. DLAD enhanced physician performance when used as a second image reader, and the algorithm improved quality and efficiency in clinical practice.
AI’s role in support of clinical decisions and evaluation of management options following diagnosis is coming fast in a field where the specialist faces a multitude of obligations. Dermatology patients often have complex medical backgrounds resulting in multiple symptoms. Therapies such as immunosuppressant medications, biologics, and chemotherapeutics must be carefully administered due to potential systemic effects. Plus, this busy profession allows little time for perusing medical journals to implement current research into care regimens.
In seconds, however, a properly trained AI network can assimilate reams of data11 from published articles, textbooks, guidelines, and narratives. A cognitive computing system that uses natural language processing to interpret the doctor’s query then searches the extensive database for related information. It then provides the physician with confidence-ranked answers.
This type of AI will greatly complement electronic health record systems, minimizing ineffective treatment and risk of adverse reactions, while allowing dermatologists to spend more time with patients.
8) Virtual nurses
A patient’s ability to control his or her environment has a positive impact on healing. In the healthcare facility, a virtual nurse assistant12 would give the patient that comfort 24/7. Using a popular virtual assistant device and app, the patient could manage in-room conditions such as lighting and bed adjustments. Information about treatment and scheduling could be requested, and the human nurse paged. Data obtained through the exchange would help staff prioritize and triage patient care.
A patient experiences a sense of isolation from hospitalization or a nursing homestay. Ease of connectivity with loved ones, such as viewing and responding to messages by voice command, may correlate to faster recovery.
AI trends in healthcare
When and how will AI reach a healthcare provider near you? That remains to be seen. However, the CMS Innovation Center challenges innovators from all sectors (not limited to healthcare) to explore how AI can improve health outcomes. The lucrative, three-stage AI Health Outcomes Challenge13 launched in June 2019 gives participants the opportunity to demonstrate how deep learning and neural networks can predict health outcomes and improve quality of care.
Artificial intelligence brings life-altering insights into a host of conditions and diseases among the most challenging for healthcare professionals. Evidence surrounding AI technology shows it can and does supplement diagnostics and human clinical decision-making.
- AI in Medical Imaging to Top $2 Billion by 2023. Signify Research. https://www.signifyresearch.net/medical-imaging/ai-medical-imaging-top-2-billion-2023/ June 11, 2019.
- AI for Imaging Analytics Intrigues Healthcare Orgs, Yet Starts Slow. HealthITAnalytics. https://healthitanalytics.com/news/ai-for-imaging-analytics-intrigues-healthcare-orgs-yet-starts-slow June 11, 2019.
- Quick Facts About ALS & The ALS Association. ALS Association. http://www.alsa.org/news/media/quick-facts.html June 11, 2019.
- Deep learning predictions of survival based on MRI in amyotrophic lateral sclerosis. Neuroimage Clinical. https://www.ncbi.nlm.nih.gov/pubmed/28070484 June 11, 2019.
- Alzheimer’s and Dementia Facts and Figures. Alzheimer’s Association. https://www.alz.org/alzheimers-dementia/facts-figures June 11, 2019.
- Artificial intelligence predicts Alzheimer’s years before diagnosis. ScienceDaily. https://www.sciencedaily.com/releases/2018/11/181106104249.htm June 11, 2019.
- Diagnosing Fractures with AI. JAMA Network. https://jamanetwork.com/journals/jama/article-abstract/2686776 June 11, 2019.
- Stand-Alone Artificial Intelligence for Breast Cancer Detection in Mammography: Comparison With 101 Radiologists. National Cancer Institute. https://academic.oup.com/jnci/advance-article-abstract/doi/10.1093/jnci/djy222/5307077?redirectedFrom=fulltext June 11, 2019.
- ACR Data Science Institute Releases Landmark Artificial Intelligence Use Cases. American College of Radiology. https://www.acr.org/Media-Center/ACR-News-Releases/2018/ACR-Data-Science-Institute-Releases-Landmark-Artificial-Intelligence-Use-Cases June 11, 2019.
- Development and Validation of a Deep Learning–Based Automated Detection Algorithm for Major Thoracic Diseases on Chest Radiographs. JAMA Network Open. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2728630 June 11, 2019.
- Artificial Intelligence for Clinical Decisions Support. MDedge Dermatology. https://www.mdedge.com/dermatology/article/173762/practice-management/artificial-intelligence-clinical-decision-support June 11, 2019.
- Digital Nurse’s Assistant. Smashing Ideas. https://smashingideas.com/work/digital-nurses-assistant/ June 22, 2019.
- CMS launches Artificial Intelligence Health Outcomes Challenge. Centers for Medicare & Medicaid Services. https://www.cms.gov/newsroom/press-releases/cms-launches-artificial-intelligence-health-outcomes-challenge June 11, 2019.