Introduction: History of AI in medicine. Artificial intelligence (AI) was first defined in the 1950s. However, several limitations in early models prevented widespread acceptance and application of the drug. In the early 2000s, many of these limitations were overcome with the advent of deep learning.
Now that AI systems can analyze complex algorithms and learn independently, we are entering a new era in medicine where AI can be applied to clinical practice through risk assessment models, diagnostic accuracy and workflow efficiency can be improved.
AIM (artificial intelligence in medicine) has evolved dramatically over the past five decades. Since the advent of ML (machine learning) and DL (deep learning), the applications of AIM have developed, forming possibilities for personalized medicine rather than algorithm-based medicine.
Predictive models can be used to diagnose diseases, predict response to treatment, and potentially develop preventative medicine in the future.
AI can improve diagnostic accuracy and efficiency in provider workflow, clinical operations, disease monitoring, and therapy, as well as procedural accuracy and overall patient outcomes. Let’s read more about History of AI in medicine.
History of AI in medicine
The conceptual foundation of AI in medicine dates back to the mid-20th century. In 1950, Alan Turing introduced the idea of machine intelligence with his seminal paper “Computing Machinery and Intelligence,” where he proposed the famous Turing Test. Around the same time, medical professionals began to see the potential of computers in healthcare.
Early artificial intelligence focused on developing machines that could make conclusions or decisions that humans could not make previously. The first industrial robot arm (Unimate; Unimation, Danbury, Conn, USA) joined the assembly line at General Motors in 1961 and demonstrated automated die casting.
A few years later (1964), Eliza was introduced by Joseph Weizenbaum. Using natural language processing, Eliza could communicate using pattern matching and substitution mechanisms to mimic human conversation (surface communication) as a framework for future chatterbots.
1966, Shakey, the “first electronic person,” was produced. Built at the Stanford Research Institute, it was the first mobile robot capable of interpreting instructions in Robotics and AI.
Despite these innovations in engineering, medicine has needed to adopt AI faster. However, this early period was essential for digitizing data, which later served as the basis for the future development and use of AIM.
1970s to 2000s
Much of this period is called the “AI winter,” indicating a period of low funding and interest that results in less significant progress. Another spans the late 1980s to early 1990s due to the prohibitive cost of developing and maintaining specialized digital information databases.
Despite the lack of general interest during this period, collaboration between pioneers in AI continued. It promoted the development of a research resource on computers in biomedicine by Saul Amarel in 1971 at Rutgers University.
Stanford University Medical Laboratory–Artificial Intelligence in Medicine, a time-sharing computer system, was built in 1973 and increased networking capabilities among clinical and biomedical researchers from several institutions. – Sponsored AIM workshop held at Rutgers University in 1975. These events represent early collaboration among AIM pioneers.
A “backward chaining” Artificial Intelligence system, MYCIN, was formed in the early 1970s. Based on patient information input from 14 clinicians and knowledge of approximately 600 rules, MYCIN can provide a list of potential bacterial pathogens and then adjust antibiotic treatment options.
Appropriate for the patient’s body weight. MYCIN became the framework for a later principle-based system, EMYCIN. INTERNIST-1 was later developed using the same EMYCIN framework and a more significant clinical knowledge base to aid the primary care physician in diagnosis.
In 1986, the University of Massachusetts dismissed DXplain, a decision support system. This program uses the listed symptoms to generate a differential diagnosis. When first removed, DXplain was capable to provide information on approximately 500 diseases.
Since then, it has spread to more than 2,400 diseases. By the late 1990s, there was a renewed interest in ML, particularly in the medical world, which, with the above technological advances, set the stage for the modern era of AIM.
The 1990s saw AI transition from rule-based systems to machine learning (ML) in medicine. This shift was driven by advances in computational power and the availability of large datasets. Machine learning algorithms could learn from data and improve over time, unlike their rule-based predecessors.
A notable example from this period is the development of neural networks for medical image analysis. These networks were used to identify patterns in medical images, such as mammograms and MRI scans, leading to more accurate diagnoses of conditions like breast cancer.
In 1997, IBM’s Deep Blue, while not directly related to medicine, showcased the potential of advanced AI. This chess-playing computer defeated world champion Garry Kasparov, highlighting AI’s problem-solving and pattern-recognition capabilities.
2000 to 2020: Fundamental Advances in artificial intelligence
In 2007, IBM created an open-domain question-answering system called Watson, which competed against human participants and the television game show Jeopardy!
In 2011, Unlike traditional systems that entangle either forward reasoning (following rules from data to results), backward reasoning (following directions from results to data), or hand-crafted if-then rules, DeepQA this technology is called natural uses. It uses language processing and various searches to analyze data on unstructured content to generate potential answers.
DeepQA technology can extract information from patients’ electronic medical records and other electronic resources to provide evidence-based medical answers. Thus, it has opened up new possibilities in evidence-based clinical decision-making.
In 2017, Bakkar etal used IBM Watson to successfully identify new RNA binding proteins altered in amyotrophic lateral sclerosis—Alexa, in 2014.
Pharmabot was a chatbot developed in 2015 to help educate pediatric patients and their parents on medication. Mindy created it in 2017 as an automated patient intake process for a primary care practice.
A convolutional neural network (CNN) is a DL algorithm applied to image processing that simulates the behavior of interconnected neurons in the human brain.
A CNN comprises several layers that analyze the input image to recognize patterns and create specific filters. The final result is produced by combining all the features through fully connected layers.
Limitations and challenges in the application of artificial intelligent systems in medical science
The application of artificially intelligent systems in any field, including healthcare, has limitations and challenges. Now is the time to change our mindset from reactive to proactive regarding declining new technologies. Here, we discuss challenges that are particularly relevant to healthcare.
Data Availability
Data collection is the first step towards building artificially intelligent systems (after problem selection and solution strategy development). Creating well-performing models relies on the availability of large amounts of high-quality data.
The data collection issue has been shrouded in controversy due to patient privacy concerns and recent data breaches by large corporations. Advances in technology have resulted in increased computational and analytical power and the ability to store vast amounts of data.
Facial recognition and gene analysis provide a way to identify an individual from a pool of people. Patients and the public generally have a right to privacy and a right to choose what data, if any, they want to share.
Model selection
With several algorithms and models to choose from, one must select an algorithm appropriate for the task. Thus, the model selection process is essential. Bias models are extremely simplistic and fail to capture directions in the dataset.
Black boxes
Artificial intelligence systems have a standing for being black boxes due to the complicatedness of the mathematical algorithms. Models need to be made more accessible and interpretable. Although there has been some recent work in this direction, some progress has yet to be made.
Creating a Biased Model Biased Data
Artificially intelligent systems are then trained with a portion of the collected data (also called the training data set), and the rest of the data is saved for testing (also called the testing data set). Is).
Thus, if the data collected is biased, meaning it targets a particular race, gender, or age group, the resulting model will be biased. The data thus collected should accurately represent the population for which it is intended to be used.
Fragmented data
Another limitation of AI applications is that models that an organization spends time and effort designing and deploying for a specific task (regression, classification, clustering, NLP, etc.)
It cannot be seamlessly transferred to another organization for use. Due to privacy concerns, data sharing between healthcare organizations is often inaccessible or restricted, resulting in data fragmentation that limits model reliability.
Data preprocessing
It is possible to create a biased model even after collecting unbiased data. The collected data must be preprocessed before being used to train the algorithm. The raw data often contains errors due to manual data entry or various reasons.
These entries are sometimes replaced by mathematical justification or simply removed. Care should be taken that data preprocessing does not result in a biased data pool.
Conclusion
AI has come a long way from the early Turing tests to its current incarnation. As we enter the “Roaring Twenties,” we are at the dawn of a new medical era as AI begins immersion into everyday medical practice.
The potential applications of AI in GI disease are currently limitless. This ranges from enhancing our diagnostic capacity in endoscopy, making endoscopy workflow more efficient, and even helping to more accurately risk gratifying patients for common GI conditions such as GI bleeding and neoplasia.
The history of AI in medicine is a story of continuous innovation driven by technological advances and a deepening understanding of both AI and medical science.
From early expert systems to modern deep learning applications, AI has the potential to revolutionize healthcare, improving diagnostics, treatment, and patient outcomes. As we move forward, carefully considering ethical and practical challenges will be crucial to fully realizing AI’s potential in medicine.
Also read:History of medicine in China; History of medicine in France; History of medicine in Egypt