Artificial Intelligence (AI), has been playing a robust and growing role in the world the past few decades. What most do not realize is artificial intelligence presents itself in many forms that impact daily life. Logging into your social media, e-mail, car ride services, and online shopping platforms all involve artificial intelligence algorithms to improve user experience. One major area AI is growing rapidly is the medical field; specifically, in diagnostics and treatment management. As there is a fear of Artificial Intelligence surpassing human tasks and ability, there is significant research as to how AI can aid in clinical decisions, support human judgement and increase treatment efficiency.
An Increased AI Presence in Healthcare
There are various magnitudes of AI in healthcare. Many times, AI utilizes a web database allowing doctors and practitioners to access thousands of diagnostic resources. As doctors have been deeply educated in their field and are current with present research, the use of AI greatly increases a faster outcome that can be matched with their clinical knowledge. Artificial Intelligence presents many fears, especially in the clinical setting, of eventually replacing or reducing the need for human physicians. However, recent research and data has shown that it is more likely this tool will benefit and enhance clinical diagnostics and decision making rather than reduce clinician need.
Many times, a patient can present multiple symptoms that can correlate with various conditions by both genetic and physical characteristics, which can delay a diagnosis. So, not only does AI benefit a practitioner in terms of efficiency, it provides both quantitative and qualitative data based on input feedback, improving accuracy in early detection, diagnosis, treatment plan and an outcome prediction.
The ability for AI to “learn” from the data provides the opportunity for improved accuracy based on feedback responses. This feedback includes many back-end database sources, input from practitioners, doctors, and research institutions. The AI systems in healthcare are always working in real time, which means the data is always updating, thus increasing accuracy and relevance. Assembled data is a compilation of different medical notes, electronic recordings from medical devices, laboratory images, physical examinations and various demographics. With this compilation of endlessly updating information, practitioners have almost unlimited resources to improve their treatment capabilities.
AI Machine Learning Provides More Targeted Diagnostics
With various amounts of healthcare data out in the field, Artificial Intelligence must efficiently sort through the presented data in order to “learn” and build a network. Within the realm of healthcare data there are two different types of data that can be sorted; unstructured and structured. Structured learning includes three different types of techniques including Machine Learning Techniques (ML), a Neural Network system, and Modern Deep Learning. Whereas, all unstructured data uses Natural Language Processing (NLP).
Machine Learning techniques use analytical algorithms in order to pull out specific patient traits, which include all the information that would be collected in a patient visit with a practitioner. These traits, such as physical exam results, medications, symptoms, basic metrics, disease specific data, diagnostic imaging, gene expressions, and different laboratory testing all contribute to the collected structured data. Through machine learning, patient outcomes can then be determined. In one study, Neural Networking was utilized in a breast cancer diagnostic process sorting from 6,567 genes and paired with texture information inputed from the subjects’ mammograms. This combination of logged genetic and physical characteristics allowed for a more specific tumor indicator outcome.
The most common type of Machine Learning in a clinical setting is known as supervised learning. Supervised learning uses the physical traits of the patient, backed with a database of information (in this case breast cancer genes), to provide a more targeted outcome. Another type of learning used is Modern Deep Learning, which is considered to go beyond the surface of Machine Learning. Deep Learning takes the same inputs as Machine Learning, but feeds it into a computerized neural network; a hidden layer that further files the information to a more simplified output. This helps aid practitioners that may have multiple possible diagnoses narrow down to one or two outcomes; thus, allowing the practitioner to make a more definitive and concrete conclusion.
Similar to the structured data processes is Natural Language Processing, which focuses on all of the unstructured data in a clinical setting. This type of data is from clinical notes and documented speech to text processing when a practitioner sees a patient. This data includes, narratives from physical examinations, laboratory reports, and exam summaries. The Natural Language Processing uses historical databases that have disease relevant keywords aiding in the decision-making process for a diagnosis. Using these processes can provide a more accurate and efficient diagnosis for a patient, which in turn saves time for the practitioner, and more importantly can speed up the treatment process. The faster, more targeted and specific the diagnosis, the sooner a patient can be on the road to recovery.
Artificial Intelligence Integrated in Major Disease Areas
With cardiovascular, neurological disorders and cancer consistently being the top causes of death, it is imperative that as many resources as possible are being utilized to aid in early detection, diagnosis and treatment. The implementation of artificial intelligence provides benefits in early detection by being able to pinpoint any risk alerts a patient may have.
One study involving patients at risk for stroke used AI algorithms based on their presented symptoms and genetic history to place them in an early detection stage. This stage was movement based, where any abnormal physical movement in the patient was documented and would trigger an alert. This trigger alert allowed for practitioners to get patients to a MRI/CT scan sooner for a disease evaluation. In the study, the early detection alert provided 87.6% accuracy in a diagnosis and prognosis evaluation. That said, the practitioners were able to implement treatment sooner and predict whether the patient had a higher possibility of future stroke. Likewise, machine learning was used in 48-hour post stroke patients gaining a perdition accuracy of 70% whether the patient may have another stroke or not.
Telehealth: Artificial Intelligence on a Smaller Scale
Although Artificial Intelligence is used for high-risk diseases and on a larger scale, telehealth tools are being implemented into homes of patients to help treat and prevent high-risk situations while reducing hospital readmissions. Telehealth tools allow for different metrics to be taken, documented and processed much like a more expansive AI machine. This equipment can notify practitioners immediately when a patient reports a high-risk variable. Early detection, faster diagnostics, and an updated treatment plan, reduce time and money for both the patient and hospital, while getting more immediate care. Artificial Intelligence is allowing for practitioners to make more efficient and logical decisions, bettering the care for patients as a whole; which in the end, is the ultimate goal.