E-Handbook: Role of AI in healthcare applications is more aggressive, invasive Article 2 of 4

AI usage in healthcare upgrades imaging, analytics, automation

Computer vision, deep learning and automation are making greater inroads into radiology, robot-assisted surgeries, virtual nursing assistants and fraud detection.

Though considered to be in its early stages of penetration into healthcare, AI nonetheless has a proven track record of supporting improved patient outcomes and efficiencies in applications ranging from radiology to emergency medical applications. AI usage in healthcare applications is most well-known in detecting abnormalities and defects in scans that may be missed by the naked eye as well as predicting oncoming illnesses ahead of actual symptoms.

Patient electronic health records for analysis and data mining are more available now than ever as more than 95% of hospitals and nearly 90% of office-based physicians are on record as adopting an EHR system, according to HIMSS. This trend opens the door for AI to play an even greater role in healthcare. Thanks to accessible data and advancements in algorithms and research, AI usage in healthcare is also being actively applied in areas such as robot-assisted surgeries, fraud detection and virtual nursing assistants.

Imaging shows life-changing results

Still, radiology leads in aggressively adopting AI due in large part to recent advancements in computer vision along with deep learning models built to better diagnose medical images. More specifically, AI is being used to deliver early predictions of Alzheimer's disease. Jae Ho Sohn -- a radiology resident at the University of California, San Francisco Medical Center -- and his team have developed an algorithm that can provide accurate predictions and assessments of whether patients will develop Alzheimer's disease through analysis of an FDG-PET brain scan.

In addition, AI can be used to analyze patient images 24/7 year-round even when unattended by a radiologist. AI usage in healthcare can also support emergency physicians by analyzing X-rays and other medical images immediately and determining with a high level of confidence whether a trauma patient has fractures. That's especially critical for hospitals and other emergency facilities with limited radiologist coverage.

EHR adoption is on the rise

At the radiology conference RSNA 2019, Chi-Tung Cheng, a trauma and emergency surgeon at Linkou Chang Gung Memorial Hospital in Taiwan, highlighted how his hospital uses AI algorithms to detect hip and pelvic fractures in patients, allowing emergency physicians to quickly treat some injuries without waiting an extended period of time for a radiologist to read the results.

AI can assist with workflow automation and analysis of patient data stored within the EHR to determine key risk factors for patients. That's especially important for hospitals that are reducing patient hospital readmissions to improve their healthcare outcomes. The LGI Flag tool by Medial EarlySign, for example, can take clinical data that includes lab results from complete blood count testing and other demographic information like patient age and gender and determine whether that patient is at a high risk of having a gastrointestinal disorder months ahead of any symptoms. Early diagnoses can help hospitals and patients avoid costly treatments later.

AI hurdles to jump

AI usage in healthcare, however, faces its own set of obstacles despite its popularity and unlimited potential. One of the first concerns confronting physicians and other clinical professionals is that AI is typically dedicated to specific applications. One size doesn't fit all. As a result, multiple problems may require multiple AI tools.

AI continues to make life-changing inroads into healthcare applications like radiology.

Integration and access to clinical data, such as medical imaging and patient records, can also be an obstacle to AI. Since AI models require data to train it, integrating an AI platform with a hospital's EHR or radiology system could be costly, making it more difficult to adopt these platforms without adequate data to train the different models. Many of the AI vendors had to resort to freely available clinical data to build their early AI models.

Patient trust can be another issue. Patients in rural areas still have reservations about the use of AI to help with their medical diagnosis due to a lack of understanding of AI's applications and its record of safety.

Patients today are surrounded by devices and software that use AI -- from natural language processing in Siri or Alexa to the algorithms used in advertisements by Facebook and other social media that target individuals based on their purchasing habits. As AI continues to make life-changing inroads into healthcare applications like radiology, greater acceptance can be expected on the way to improved patient outcomes and earlier predictions of diseases.

Dig Deeper on Healthcare IT systems and applications

Cloud Computing
Mobile Computing