Image by the newsweek on their website(https://www.newsweek.com/2017/10/20/using-artificial-intelligence-find-cancer-cures-682477.html)
Whenever we (or our loved ones) contract a terminal illness, the most common question is: “how long do I have to live?” This question can take various forms depending on who is asking it. It may go like, “how long will the entire treatment-plan take before remission?” To answer these questions, the health providers will base their response from a previous patient or group of patients who presented with a similar condition and disease stage. This response unfortunately is not based on statistical inference but on intuition. To objectively respond, a predictive model should be adopted.
Predictive modeling is the process of developing an algorithm (model) in an understandable way that quantifies the degree of accuracy in future (yet-to-be-seen) data. Because decisions about the future are made based on the information available; this information could be objective tangible data (disease diagnosis, staging, treatment plan) or intuition and past experience. Information technology tools such as search engines sift through data by looking for patterns that are relevant to the problem at hand and return answers. The process of developing these tools can be called machine learning, pattern recognition, data mining, predictive analytics and knowledge discovery. The ultimate objective is to make an accurate prediction. The most common term used to best describe all the information technology techniques involved in prediction and pattern recognition is artificial intelligence.
Artificial Intelligence (AI) aims at mimicking human cognitive functions. The adoption of AI in healthcare has increased due to availability of more healthcare data and rapid progress in analytical techniques. Healthcare data could either be structured or unstructured.
Artificial intelligence can assist healthcare staff in better decision making. Although AI cannot replace human judgement entirely; in certain functional areas such as radiology and pathology, it can improve or limit on human judgement.
The four questions that result from the discussions on big data and analytics in healthcare are:
- What is the motivation of applying AI in healthcare?
- What are the healthcare data types analyzed using AI?
- What are the existing mechanisms that enable AI systems to generate clinically meaningful results?
- What are the disease types being currently tackled by the artificial intelligence community?
Motivation for the application of AI in healthcare
AI uses sophisticated algorithms to “learn” features from large volume healthcare data and uses the generated insights to influence clinical decision making. AI can also be equipped with learning and self-correcting abilities that can improve its accuracy based on user feedback. These systems provide up to date medical information from research journals, textbooks and clinical guidelines to inform patient care. AI can also improve on therapeutic and diagnostic errors inevitable in human clinical practice, by extracting useful information from large patient populations. Artificial intelligence can assist in making real-time inferences on health risk alerts and health outcome prediction.
Healthcare data types analyzed by artificial intelligence
AI systems need to be trained through data generated from clinical activities (screening, diagnosis, treatment and outcomes) before they can be deployed in healthcare applications. This enables them to learn from similar groups of subjects, the association between subject features and clinical outcomes of interest. These healthcare data could include patient demographic characteristics, medical notes, electronic recording from medical devices, physical examination data, clinical laboratory reports, radiological and pathological images.
During the diagnostic process, AI analyzes data from diagnostic images (X-rays, MRIs, CT-scans and echo-cardiograms) and test reports from laboratory and other medical equipment (blood pressure, thermometers and pulse oximeters). With the wide adoption of genetic analysis and testing of patient genomes, AI provides a good opportunity to analyze DNA and RNA sequence data.
AI applications often begin by converting unstructured text information to machine understandable electronic medical records.
Mechanisms that enable AI systems to generate clinically meaningful results
AI systems adopt classical machine learning (ML), Natural Language Processing (NLP) and Deep Learning (DL) techniques. Machine learning clusters patient traits to infer the probability of patient outcomes while natural language processing analyzes unstructured data (clinical notes, medical journals, treatment guidelines) by converting it into machine readable structured data that can be analyzed using machine learning.
These techniques can only be powerful if they are motivated by clinical (healthcare) problems and are applied to assist clinical practice (and public health initiatives) in the end.
Disease Types Focus of Artificial Intelligence
There are a myriad of diseases and disease categories. Artificial intelligence efforts in healthcare have focused on three key disease areas: Cancer (Neoplasms), Heart (Cardiovascular) and Nervous (neural) diseases.
In cancer, the IBM Watson for oncology (WFO) was demonstrated by Somashekhar’s team at Memorial Sloan Kettering Cancer Centre in New York, to assist in the diagnosis of various cancers. Andre Esteva and Sabastian Thrun at Stanford University demonstrated the role of deep neural networks in the diagnosis of skin cancer.
For nervous diseases, AI has been used by Chad Bouton and colleagues at Battelle Memorial Institute – Columbus, Ohio to restore and control movement in patient with quadriplegia. Dario Farina’s team at Imperial College London described the power of man-machine interface using discharge timings of spinal motor neurons to control upper limb prostheses.
The potential of applying AI systems in diagnosing heart diseases through cardiac image analysis has been discussed at length by various groups such as those of Dilsizian and Siegel at the University of Maryland School of Medicine in Baltimore.
The bias in these three disease areas could be attributed to their high disease burden (morbidity) and death (mortality) in many regions globally.
Specifically, artificial intelligence can be used in pattern recognition during differential diagnosis, radiology and laboratory procedures. Many diagnostic and therapeutic medical equipment are robotic with in-built software that rely on conditional (if…then) statements. Unless certain conditions are met, the procedure won’t proceed to the next stage. This greatly minimizes on human errors.
Many public hospitals suffer from insufficient stocking of medicines and other medical supplies. With predictive modeling, it not only possible to properly manage the inventory but properly plan for treatment stocks by predicting drug resistance and demand.
In pharmaceutical industries, there is a bank of potential chemical molecules that could be used in the development of particular medicines. With artificial intelligence, it is easier to identify potential medicines through high throughput screening (HiTS). This can be applied to new or existing medicines with new indications (use of that drug for treating a particular disease).
The medium-term plan III (MTP3) of Kenya’s vision 2030 has adopted Universal Healthcare Coverage (UHC) as one of its four agenda. Artificial intelligence can be used when choosing the best payment models for various public health institutions, categorization of health insurance policy categories, reduction of fraud in national health insurance schemes and stratification of health facilities based on resource-needs.
Clinical Research among other things aims at determining patient and treatment outcomes. By determining clinical outcomes, it is possible to predict survival rate at 1, 3, 5 or ten years.
So next time before asking about how long you have left; could you ask the healthcare practitioners about their adoption of big data analytics and artificial intelligence?
By: Gerald Omondi.