Application of Artificial Intelligence Methods to Diabetes

In This Article, we are going to see about artificial Intelligence methodologies and their application to diabetes in the most efficient way. Let's get into the Article.

 

Abstract

Over the last ten years, the combination of continuous glucose monitoring and data from insulin pumps has changed the management of diabetes. More recently, wristbands or watches have been able to track a wide range of physiological characteristics and functions, including heart rate, sleep duration, steps taken, and activity. Future updates will include more information including barometric pressure, hydration, and geolocation. When all of these factors are considered, it can assist patients and clinicians make decisions. In recent years, there has been a rising interest in the development and implementation of artificial intelligence (AI) technologies to decision assistance and knowledge acquisition. Similar new scenarios have emerged in the majority of medical professions. Diabetes-related journals are increasingly including articles that discuss the use of AI techniques in the treatment of the disease. In conclusion, diabetes management situations have undergone a significant transition that compels diabetologists to draw on expertise from other fields. The purpose of this essay is to clearly explain the most popular AI approaches in order to encourage the involvement of healthcare professionals—doctors and nurses—in this field.

 

AI Methodologies

 

Expert Medical Systems

  • Expert systems (ES) are the most prevalent class of AI systems used in everyday clinical practice. In order to assist healthcare providers in their daily job, they are defined as systems with the capacity to capture expert knowledge, facts, and reasoning procedures. 
  • By using inference techniques to support decision-making or problem-solving, ES attempts to emulate the clinical competence of the doctor. ES is capable of handling facts to draw logical judgments. Among its many utilities, ES is used for picture interpretation, diagnosis support, and alert creation.

 

RBR

  • RBR relies on the transmission of knowledge from a subject-matter expert to a computer. As a result, the computer must be able to tackle issues that would often require a specialist. If-then sentences are used to describe knowledge so that the chain of reasoning can be clarified. 
  • A series of conversations between the expert and the knowledge engineer who will ultimately build and test the ES serves as the beginning of the knowledge acquisition process. The domain expert sets all the possibilities during these interviews, and the engineer then encodes this information to make it "machine-interpretable."

 

CBR

  • CBR uses previously effective solutions to related problems to find answers to new ones. Features of case studies must be mentioned in order to be useful in obtaining other cases. 
  • Additionally, features must be discriminatory enough to prevent the retrieval of case studies that might provide incorrect solutions due to their extreme differences. Unlike RBR, CBR only needs to find new examples with important features—this is how CBR "learns"—instead of requiring an explicit domain model.

 

Machine Learning

  • Algorithms that use machine learning (ML) have the capacity to learn over time without being explicitly designed. Problem-solving, typically based on data classification, is one of machine learning's key aspects. Heuristic methods have been gradually replaced by ML techniques. 
  • Data mining is the process of extracting useful knowledge from huge databases, like those found in electronic medical records, that may contain implicit regularities. 
  • Additionally, ML can be used in fields where a computer program must dynamically adjust to changing circumstances. For instance, ML algorithms are important in an artificial pancreas system to learn from each patient monitoring data set and adapt over time.



ANN

  • ANNs are based on how linked neurons work in the human brain. The basic unit, the neuron, produces only one output while taking in several inputs. Each link has a weight that corresponds to the significance of the output. 
  • The neural network "learns" by practicing with predetermined inputs, comparing the observed output to the expected one, and using the difference in output to change weights. As a result, the links that lead to the right answers are reinforced, while the links that lead to the wrong responses are weakened.

 

Deep learning

  • A new area of machine learning called deep learning is based on how neurons behave in human brains. It uses a hierarchical level of ANN to carry out the classification process, hence it may be seen as an evolution of ANN. 
  • Deep learning algorithms are especially effective at learning processes and give systems based on them a high level of intelligence. The term "deep" in deep neural networks alludes to the fact that numerous layers of processing convert input data—whether it be images, audio, or text—into an output that can be used to make judgments.

 

Diabetes and AI Technologies: Practical Applications

Patient Decision Support With CBR

  • Researchers from London's Imperial College have conducted one of the most pertinent experiences on the application of ES to patient decision assistance. They have created and tested a CBR-based bolus calculating algorithm. 
  • This technology, which is integrated into the patient's smartphone, makes use of data from continuous glucose monitoring. The advantages of this tool over standard bolus calculators have been demonstrated in a pilot feasibility study that has been published.

 

Systems with Closed Loops In light of FL

  • FL-based algorithms have been utilized successfully for closed-loop investigations, especially in the ambulatory context, in addition to proportional derivative integral (PID) and model predictive control (MPC).
  • Mauseth et al. published one of the earliest works that included FL for closed-loop systems in 2010. Both the BG and the rate of change of glucose input to the controller. The system assigned a coefficient using a matrix, and after defuzzification, suggested an insulin micro bolus. Three years later, the technique was successfully tested in a pilot study.

 

Computer Interpretable Guidelines (CIGs) for the Management of Gestational Diabetes

  • Clinical practice recommendations are valuable tools for raising the standard of treatment. Tools for supporting decision-making can be created through formalization as CIGs using a sophisticated RBR system.
  • In another article in this special section of the journal, clinical experience with gestational diabetes CIGs used for patients' and doctors' decision assistance is presented.
  • Briefly stated, a pilot study revealed that, compared to standard therapy based on in-person visits, patients reported greater levels of patient satisfaction and blood glucose monitoring compliance.

 

Retinal disease detection With ANN

  • Deep learning ANN has recently demonstrated its ability to accurately and quickly detect diabetic retinopathy or diabetic macular edema in retinal fundus images.
  • 16 The scientists have created an algorithm that determines the degree of diabetic retinopathy based on the brightness of the pixels in a fundus image. Large datasets of images were used to train the function, which was then assessed at a high specificity operating point and a high sensitivity operating point to achieve very high results.

 

Conclusion

A process of adaptation in the field of diabetology is necessary to include new strategies for managing diabetes. For both patients and healthcare professionals, technology, in particular sensors and computer programs, has emerged as a crucial tool in the management of diabetes. Doctors and nurses must overlook the fundamentals in order to better identify answers to each patient's circumstances, even though modern diabetes care units should have a diabetic technologist to deal with technology. In addition to a list of pertinent papers on AI used to treat diabetes, this article offers a comprehensive explanation of the fundamental ideas, definitions, and terminology typically used in applications linked to Artificial Intelligence.