Integrating Artificial Intelligence into the Practice of Transfusion Medicine

Abstract

OVERVIEW OF ARTIFICIAL INTELLIGENCE Artificial intelligence (AI) refers to computer systems capable of performing complex tasks that historically only a human could do, such as reasoning, making decisions, or solving-problems. John McCarthy first described the term as early as 1956, as the science and engineering of making intelligent machines. AI and machine learning (AIML), an offshoot of computer science, are rapidly evolving in various fields of healthcare. Today, AIML is used to describe the use of computers and technology to simulate intelligent behavior and critical thinking akin to a human being.[1] AI is an umbrella term that encompasses a wide variety of technologies, including machine learning, deep learning, and natural language processing. Machine learning may use a flowchart-based algorithmic approach or a database approach. Google Translate uses deep learning algorithms to translate text from one language to another. ChatGPT uses large language models to generate text in response to questions or comments.[2] AI has the potential to be applied in almost every field of medicine, be it in radiology for interpreting X-rays or computed tomography scans or in robotic surgery for performing complex surgeries guided by a surgeon. AI is also used to automate appointments, online check-in of patients at outpatient departments, digitization of medical records, and so on.[3] A computer may use a flowchart-based approach as with patient history taking, to narrow down the diagnosis or possibilities using a large amount of data, fed into machine-based cloud networks (neural networks). Artificial neural networks are organized in several layers to imitate how the human brain interprets and draws conclusions from information. AI may also use a database approach as with radiology films which are based on deep learning or pattern recognition. AI applications have transformed health care the world over, particularly in the field of medical imaging and diagnostics.[3,4] Unknown to us, AI has been in use in our field of immunnohematology and transfusion medicine. This editorial is aimed to explore areas where AIML can be employed in transfusion medicine. ARTIFICIAL INTELLIGENCE IN TRANSFUSION MEDICINE –SUBSPECIALTIES Artificial intelligence in donor registration and counseling The donor selection deferral criteria are so vast and ever-changing that it is nearly impossible to remember all of them. Human intelligence is limited and often needs assistance from books, documents, or Google searches to accept or defer a donor. This is where AI is helpful. The amount of information that an AI tool can remember is many times more than the human mind. Once the selection deferral criteria are fed in or access to appropriate documents specified, AI-operated systems can answer any donor query in a standardized manner that too with scientific reference. Donor counseling involves complex topics such as exploring donor’s sexual health, his habits such as intravenous drug abuse, and answering doubts or queries that the donor may have. However, donor queries are fairly repetitive such as – “I have diabetes or high blood pressure can I donate?” and “I am on medication for hypothyroidism can I donate?” These largely repetitive queries can probably be better handled by the AI system once programmed. For instance, “Is the process of blood donation painful?” can be answered differently by different blood centers, by different staff in the same blood center, or by the same staff on different occasions. This is because human intelligence is dependent on our memory, mood, energy levels, and time at our disposal. All these can be overcome and our replies can be standardized using an AI tool. While answers to all questions may not be available on Day1, as the tool learns more and more or gets access to good reliable documents, it can possibly give out standardized replies with no dependence on memory or mood. Such tools can be linked with PubMed to source authentic documents, which may however entail substantial fees. ARTIFICIAL INTELLIGENCE IN DONOR RECRUITMENT AND DONOR RECALL Recall of deferred donors or recruitment of new donors after answering their doubts can be a herculean task and many large blood centers employ a lot of call center staff to do specifically this. However, the repetitive nature of the job makes it uninteresting for the ordinary human mind. In case of AI, there is no fatique. It can answer questions the same way be it noon or midnight. It can send out reminder messages with ease, as configured, eliminating the need for human interference. People employed in this area may be utilized better to provide more and more information to the AI tool and improve machine learning or alternately these people could spend more time with actual donors, adding a personal touch to blood donation. Caring with a personal touch is possible only with humans not with machines. Analysis of donor deferral or other quality indicators with reasons and percentages can be done in a jiffy using an AI tool. The number of call center staff or time spent by employees for quality analysis can markedly come down if AI is employed for these activities. Artificial intelligence in blood donation If robots can perform surgeries, why cannot robots collect blood from a donor? Technically feasible, but the additional cost incurred in comparison with the salary of a phlebotomist is probably going to make this unviable until another 25 years when Asian countries will reach a level where phlebotomist salary will begin to reach higher levels than the cost of a robot. Even if that does happen, you will still need human being to supervise the robots. Artificial intelligence in hemovigilance Analyzing adverse donor or recipient reactions can be easily done using flow chart-based approach based on symptoms and signs to arrive at the cause of the reaction. Alternately one can also use database approach to make the computer learn that certain groups of symptoms represent certain reactions. Given that nearly all reactions fit into a certain category, specified by the International Society of Blood Transfusion, this is easier to implement by means of an app that uses in built information and gives out a standard diagnosis using inclusion and exclusion criteria. Again caring for a donor or patient with a reaction can only be done by humans and not machines. AI can help diagnose but not treat the donor or the patient. However, the AI tool can call each and every donor, postdonation, and get postdonation feedback which would be difficult for us. The use of AI for big data analysis comprising of data in National Hemovigilance Programs will help governments to govern hemovigilance programs better. Artificial intelligence in immunohematology Immunohematology (IH) was probably one of the earliest fields in medicine where AI was employed. The image of the reaction in gel cards or microplates is captured, read, and interpreted, and the blood group is determined using pattern matching. However, any discrepancy and the machine give the result as “INVALID” and ask you to interpret yourself. This is where machine learning comes in. A database approach coupled with the principle of deep learning or pattern recognition can be used here. This involves teaching a computer through repetitive algorithms to recognize not only the common blood group patterns but also rarer ones and prompt the blood center to carry out additional tests for more accurate analysis. By feeding the database with hundreds or thousands of patterns, one can probably interpret the blood group as accurately as using human intelligence. Likewise interpreting clinically significant antibodies from an 11- or 16-cell panel can be done by pattern matching instead of laboriously ruling out cell by cell manually. These things are already in use in AutoMax by Tulip, NEO microplate system by Immucor, or orthovision by ortho clinical diagnostics to a variable extent. The full potential of AI in IH is yet to be explored. Automation in IH has already reduced the burden on paramedical staff and this is here to stay and grow. Artificial intelligence in component separation Based on the size of your inventory, the AI tool can tell you accurately how much platelets you need to prepare or how much cryo. It should also be possible to tell you through pattern recognition or based on weight which are the nonconforming units in stock which need your special attention. These are tasks difficult to do consistently by mortal human beings. Artificial intelligence in TTI testing Computerized systems to interpret results of enzyme-linked immunoassay, chemiluminescence, or nucleic acid testing have been around for more than 2 decades. No great intelligence is involved here, except for simple mathematical calculations. AI is a branch of computer science capable of analyzing complex medical data and establish meaningful relationships within a data set or between data sets. AI would perhaps be helpful in big data analysis such as analyzing and comparing the TTI rate across 4000 blood centers of India or comparing the TTI rate of 1 blood centers with another or one country with another.[5] Artificial intelligence in blood transfusions RFID techniques and barcoding are already in use for patient and blood bag identification. AI systems linked to cardiac monitors can detect changes in vital signs – temperature, heart rate, blood pressure, and oxygen saturation, and help prevent an impending adverse reaction. Artificial intelligence in inventory management These are easy things to do. By linking the AI tool to the hospital information system that contains, blood requests or data on patients likely to need blood, derived from history or laboratory tests, AI can determine blood needs for the day, week, or month. It can tell which are the patients for whom blood must be issued and for whom it can be postponed based on linkage with the laboratory information system. BENEFITS, LIMITATIONS, AND THE SCOPE FOR FUTURE While it is clear that AI has several benefits in our field waiting to be exploited, it comes with some challenges. With start-ups trying out a new tool every day, it is difficult to safely implement the same without proper validation. People who are involved in developing AI tools should first get a basic knowledge of IH and transfusion medicine. Further, we probably need regulatory tools or guidelines for testing the safety of an AI tool. Like any emerging technology, AI also needs continuous monitoring to ensure that it does not harm the patients or the community. While harnessing the benefits of AI, it is equally important to understand its limitations.[6]

Ähnliche Arbeiten