Artificial Intelligence (AI) is a field within computer science that has wide applications and has transformed medical technologies. It is often considered the branch of computer science that can handle complex problems with minimal theory and multiple applications. AI is used to assist researchers in analyzing large data sets, enabling precision medicine and assisting clinicians in improving patient outcomes. New techniques in AI can bring together different types of data to understand new information obtained from multiomics datasets. Analysis of high-quality data combined with machine learning, a subset of AI, can help modify unhealthy behaviors of patients, predict the risk or recurrence of chronic diseases after surgical and curative treatments, predict the progression and survival rates of patients with chronic diseases, identify therapeutic need, create better clinical trial interpretations, and identify new targets. However, to effectively implement precision medicine in healthcare, a more user-friendly interface will be required. If AI technologies are applied correctly, fairly and robustly in close collaboration with human intelligence, it is expected to open up new possibilities for effective and personalized healthcare services worldwide. In this review, the general outline of AI technology, its application areas in healthcare and its future are overviewed.
1. AI in Pathology Image Processing
Digital pathology is becoming an increasingly important technological requirement in the laboratory setting and plays a vital role in contemporary clinical practice. Histopathologists can now handle digital slide photos with much more ease and flexibility than they could ten years ago. They can also exchange images for telepathology and clinical use due to advances in processing power, much faster networks and less expensive storage. Whole slide imaging (WSI), which enables entire slides to be photographed at high resolution and saved permanently, has evolved during the last 20 years in pathology digital imaging. The FDA has approved the WSI system of digital pathology, ushering in a new era for digital image analysis in the field. Artificial Intelligence (AI) has so far been used primarily in radiology and cardiology for image-based diagnosis.
There is a growing area of active research on its application in pathology, with many research organizations and committed businesses. Images produced by WSI are an abundant source of information; their size (100k x 100k is not uncommon) makes them more complex than many other imaging techniques; they also have color information (H&E and immunohistochemistry); there is no obvious anatomical orientation as in radiology; information is available at multiple scales (e.g., 4x, 20x); and there are multiple z-stack levels (each slice has a finite thickness and depending on the plane of focus, this will result in different images). It is clear that a normal reader cannot possibly extract all the visual information.
WSI is currently in use for training at tumor boards, conferences, online seminars, and presentations. With WSI capabilities, AI tools can aid in the continuing education of the upcoming generation of pathologists by offering standardized, interactive digital slides that are available for multiple users to share anytime, anywhere. Based on histological features, Artificial Intelligence (AI) can be used to predict prognosis and treatment outcomes. Images of various tumor characteristics, surrounding microenvironment, and genetic data can provide critical information in a concise way by directly linking them with survival outcomes as well as response to adjuvant/neoadjuvant therapy. Wang et al. used H&E stained tissue microarray slides to build a machine learning model that predicted recurrence in early-stage non-small cell lung cancer (NSCLC) based on nuclear orientation, nuclear size, texture, and tumor architecture. In two validation cohorts, their model’s prediction proved to be an independent predictive factor for recurrence prediction with 82% and 75% accuracy, respectively.
In 2018, Saltz et al. drew attention to the prognostic implications of Artificial Intelligence (AI) tools. They employed a convolutional neural network to enhance pathologist response to the automatic identification of the spatial organization of tumor-infiltrating lymphocytes in images obtained from the Cancer Genome Atlas. According to their research, this feature can predict the course of 13 different cancer subtypes. Similar research by Yuan et al. provided a model for examining the spatial distribution of lymphocytes in relation to tumor cells on triple-negative breast cancer white matter irradiation. They discovered a clear relationship between the spatial distribution of immune cells and late recurrence in ER-positive breast tumors, in addition to identifying three different types of lymphocytes.
2. Commentary on Biochemical and Clinical Tests with Artificial Intelligence
Artificial intelligence (AI) is also rapidly entering the world of health science applications. The most important areas are biochemical and clinical tests. Because both patients and physicians always expect to receive the most accurate and intuitive results from laboratories, the hand of AI can provide an opportunity to not only speed up the process of obtaining results but also eliminate erroneous laboratory results. Moreover, Artificial Intelligence (AI) is also being used in clinical trials for proper selection of samples and detection of early signs of adverse effects or toxicity.
Biochemical tests are exhibiting great diversity and there are many parameters that can be successfully handled by Artificial Intelligence (AI) software. For example, attempts have been made to commit oxidative stress parameters and anti-oxidative capacity to predict certain neurodegenerative diseases such as Alzheimer’s and Parkinson’s diseases. In clinical trials, as we have mentioned that AI can be very useful to consider clinical testing right at the beginning, there have also been a lot of ventures to be able to use AI for interpretation of biochemical or clinical test results. For example, according to one of the previous studies, software algorithms are providing faster and more accurate diagnoses than pathologists themselves.
3. Artificial Intelligence in Precision Medicine and Genomics
Precision medicine, also known as personalized medicine, is an innovative approach to preventing or treating diseases by taking into account differences in an individual’s genetic history, environment, and lifestyle. Precision medicine recognizes the important fact that not all patients respond in the same way to the same treatment. It takes a patient-centric approach by analyzing clinical, genomic, and pharmacogenomic data rather than a symptom-centric approach. In the traditional healthcare system, physicians plan treatments based on symptoms. Since symptoms can vary greatly across individuals, genomic, metabolic, and clinical data should be used together to create a more personalized treatment plan. In this way, the quality of healthcare services can be improved through the application of a personalized precision medicine approach rather than the symptom-based approach of the traditional healthcare system.
Artificial intelligence can be used in medicine in two different ways: virtual and physical. Virtual uses of Artificial Intelligence (AI) include applications ranging from electronic health records to neural networks that guide patient treatment. Physical machines, such as artificially intelligent prosthetics for the disabled and robots that assist in surgery, are physical subsets of artificial intelligence. The most common applications of precision medicine are genetic screening for disease prediction and diagnosis, and pharmacogenomics for drug response prediction. Artificial intelligence and machine learning techniques have been shown to be useful in calculating genetic risk for diseases and determining ‘polygenic risk scores’ to identify individuals at high genetic risk for certain diseases. Predictive algorithms can identify disease clusters not recognized by physicians and guide the selection of personalized treatment options for these patients. Another option is to continuously monitor people with a genetic predisposition to disease, allowing early diagnosis at the onset of disease. This would avoid the need for complex treatments. Such monitoring could be further developed by implementing artificial intelligence techniques in a new generation of sensors.
Genomic studies and next-generation sequencing have progressed rapidly since the first description of DNA by Watson, Crick and Franklin in 1953. Sequencing technologies have reached a stage where the entire genome can be sequenced in a day. While whole genome sequencing (WGS) covers the entire genome, whole exome sequencing (WES) focuses only on protein-coding regions and both produce massive amounts of genomic data to analyse. These are important to help scientists understand how genetic variation is associated with a disease by affecting important cellular processes such as cell growth, cell differentiation, metabolism and DNA repair. Several deep learning models have been developed to analyse large genomic datasets and identify genetic variants within the whole genome, for example DeepVariant is an analysis pipeline, a deep convolutional neural network model. DeepVariant can call genetic variants from next-generation DNA sequencing data, enabling patient stratification based on statistically significant variants associated with disease phenotypes.
4. Artificial Intelligence in Precision Medicine and Genomics
Precision medicine, also known as personalized medicine, is an innovative approach to preventing or treating diseases by taking into account differences in an individual’s genetic history, environment, and lifestyle. Precision medicine recognizes the important fact that not all patients respond in the same way to the same treatment. It takes a patient-centric approach by analyzing clinical, genomic, and pharmacogenomic data rather than a symptom-centric approach. In the traditional healthcare system, physicians plan treatments based on symptoms. Since symptoms can vary greatly across individuals, genomic, metabolic, and clinical data should be used together to create a more personalized treatment plan. In this way, the quality of healthcare services can be improved through the application of a personalized precision medicine approach rather than the symptom-based approach of the traditional healthcare system.
Artificial intelligence can be used in medicine in two different ways: virtual and physical. Virtual uses of Artificial Intelligence (AI) include applications ranging from electronic health records to neural networks that guide patient treatment. Physical machines, such as artificially intelligent prosthetics for the disabled and robots that assist in surgery, are physical subsets of artificial intelligence. The most common applications of precision medicine are genetic screening for disease prediction and diagnosis, and pharmacogenomics for drug response prediction. Artificial intelligence and machine learning techniques have been shown to be useful in calculating genetic risk for diseases and determining ‘polygenic risk scores’ to identify individuals at high genetic risk for certain diseases. Predictive algorithms can identify disease clusters not recognized by physicians and guide the selection of personalized treatment options for these patients. Another option is to continuously monitor people with a genetic predisposition to disease, allowing early diagnosis at the onset of disease. This would avoid the need for complex treatments. Such monitoring could be further developed by implementing artificial intelligence techniques in a new generation of sensors.
Genomic studies and next-generation sequencing have progressed rapidly since the first description of DNA by Watson, Crick and Franklin in 1953. Sequencing technologies have reached a stage where the entire genome can be sequenced in a day. While whole genome sequencing (WGS) covers the entire genome, whole exome sequencing (WES) focuses only on protein-coding regions and both produce massive amounts of genomic data to analyse. These are important to help scientists understand how genetic variation is associated with a disease by affecting important cellular processes such as cell growth, cell differentiation, metabolism and DNA repair. Several deep learning models have been developed to analyse large genomic datasets and identify genetic variants within the whole genome, for example DeepVariant is an analysis pipeline, a deep convolutional neural network model. DeepVariant can call genetic variants from next-generation DNA sequencing data, enabling patient stratification based on statistically significant variants associated with disease phenotypes.
5. Machine Learning in Pharmacogenomics
Pharmacogenomics (PGx) involves understanding how a patient’s genetic profile affects their response to medications, predicting how an individual metabolizes medications and potential side effects. It is well known that genetic variations can affect drug response, particularly variations in genes involved in drug absorption, distribution, and metabolism, all of which affect the pharmacodynamics and pharmacokinetics of a drug. The aim of pharmacogenomics is to prescribe the most effective drug at the correct dose, reduce the risk of side effects, increase treatment efficacy, and enable personalized medicine by identifying a patient’s genetic variations. The FDA issued a guidance in 2013 for the pharmaceutical industry and researchers engaged in drug development, providing recommendations on when and how genomic information should be used.
It is recommended to implement pharmacogenomics evaluation in early-stage clinical trials to identify populations based on genetic influence on drug exposure, dose response, common adverse reactions and initial efficacy that should receive lower or higher doses of the drug or longer titration intervals. Although this approach has not yet been widely adopted by pharmaceutical companies, the application of artificial intelligence methods for patient stratification using clinical and genomic data is emerging and is expected to grow rapidly. Patient stratification involves the complex integration of heterogeneous sociological, demographic and biomedical data to classify patients into subpopulations for clinical practice and clinical trial design. Electronic health record-linked DNA biorepositories have found successful use in predictive modeling with the integration of pharmacogenomic and sociometric data such as gender, age, etc., to identify better treatment options.
Here are some examples of recently published open source deep learning software that apply artificial intelligence in patient stratification and healthcare coordination: DeepPatient is an unsupervised deep learning method for enhancing clinical decision systems and obtaining patient representation. Hierarchical regularities and dependencies have been recorded in the aggregated electronic health records of nearly 700,000 patients. DeepPatient can predict the probability of developing various diseases in patients. DeepCare is another deep dynamic memory model for predictive medicine. This model uses electronic health record data including drug codes, diagnoses, and procedures to predict unplanned readmissions and high-risk patients for mental health and diabetes patient groups.
6. Drug Discovery and Repurposing
Drug design is generally recognized as a distinct phase in the drug discovery process. It focuses on the development, optimization, and refinement of potential drug compounds. Drug repurposing is a faster and more cost-effective process than developing new drugs from scratch. It focuses on discovering novel pharmaceutical uses for drugs that were originally developed for specific medical indications. In silico studies include a number of tasks performed on computers to aid drug screening, drug design, and repurposing through investigating interactions between targets and drugs. Computational drug design is not a new concept. However, with advances in hardware and software, the use of computational approaches, Artificial Intelligence (AI), and machine learning models has grown rapidly.
Deep learning models, including neural networks, have been developed to study drug-drug interactions, drug-target interactions, protein-protein interactions, DNA-protein interactions, and investigation of disease mechanisms. Experimental methods to study these interactions are labor intensive, time consuming and expensive. Artificial Intelligence (AI) models have great potential to reduce the time and cost required for such analysis and subsequent drug discovery. Studies focusing on drug-drug interactions try to understand how a drug works when it is given together with another drug and how this may change the way the drug works. A detailed investigation of all the AI based software and tools used for drug discovery is beyond the scope of this review. However, we can give Google DeepMind, DeepChem, AlphaFold2 developed by DeepBind as examples of the most commonly used Artificial Intelligence (AI)-based software for drug development, discovery and analysis. Interested readers are referred to the excellent review of Qureshi et al.
Concluding remarks and future perspectives
Healthcare is moving towards a more personalized and targeted approach to diagnosis, treatment and prevention. Therefore artificial intelligence, clinical genomics, big data and pharmacogenomics are crucial for the future development of precision medicine. By harnessing the power of genomic and molecular data, precision medicine will help healthcare professionals and researchers access large amounts of medical data and make more accurate diagnoses. Currently there is no system that can simultaneously compare multi-omics data to predict more accurate and personalized outcomes. To effectively implement precision medicine in healthcare, a more user-friendly interface will be required. If Artificial Intelligence (AI) technologies are applied accurately, fairly, and robustly in close collaboration with human intelligence, it is expected to open up new possibilities for effective and personalized healthcare services across the globe.
Read Also:
-
- Artificial Intelligence (AI) In Diagnosis And Treatment
- Findings And Discussions On Artificial Intelligence (AI) In Healthcare
- Artificial Intelligence (AI) (Online Resource): The Panacea For SMEs In Healthcare
- Main Use Cases Of Artificial Intelligence (AI) In Healthcare
- Ethical, Legal And Other Challenges In The Use Of Artificial Intelligence (AI) In Health Care
Leave a Reply