Artificial Intelligence and Machine Learning in Endocrinology and Metabolism: The Dawn of a New Era
Open Access
- 28 March 2019
- journal article
- editorial
- Published by Frontiers Media SA in Frontiers in Endocrinology
- Vol. 10, 185
- https://doi.org/10.3389/fendo.2019.00185
Abstract
The rapid growth of technology in the past couple of decades has paved the way for development of novel techniques that can solve scientific questions at a rate that is far beyond the capability of humans. One such example is the field of Artificial Intelligence (AI) and Machine Learning (ML). AI is a discipline that deals with the study and design of intelligent agents, that is, devices that intricately perceive their environment and take actions that maximize the chances of achieving their goals (1). AI, in a way, mimics the structure and operating methodologies of a human brain (2). AI has two forms of application: physical and virtual (3). The physical component is mainly represented by robots. Derived from a Czech word robota, meaning “forced labor,” the physical robotic forms were conceptualized by inventors such as Leonardo Da Vinci (3). This component has been widely used in the field of endocrinology, such as robot-assisted surgery of adrenal or prostate cancer. Examples of virtual applications of AI are electronic medical records (EMR), where specific algorithms are used to identify subjects, and harness health related data (3). ML is a field of AI that deals with the development of models and intricate networks that enable computer systems to improve their performance on a specific task progressively (4). ML algorithms can be: (i) unsupervised (spontaneous pattern detection), (ii) supervised (building algorithms based on prior examples), or (iii) reinforcement learning (utilization of reward/punishment techniques to obtain the desired result) (3). A common use of ML in daily life includes flagging spam in an e-mail, autonomous driving and selecting the best route for daily commute. In the field of medicine, AI/ML technology can have substantial impact at three levels: physicians, by improving the diagnostic accuracy and assisting with therapeutic and surgical interventions; health systems, by enabling improved workflow and reduction in errors; patients, through tailoring of diagnostic, and treatment modalities based on the unique phenotypic and genetic features of individual patients (5). In this review, we focus on the virtual components of AI and ML and provide some examples for the utility of AI/ML in endocrinology and metabolism. From early ML tools like logistic regression which found their utility in medicine several decades ago, AI/ML methods have become far more multifaceted and have revolutionized the field of medicine through their ability to compute and analyze vast and complex array of datasets which would not be feasible solely with trained human skillsets (2). Several AI/ML methods have proven their utility in the diagnosis and management of various endocrinopathies. Gradient forest analysis, a ML technique, was applied in a study to identify factors contributing to variation in all-cause mortality among subjects in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial (6). This technique detected four risk groups based on hemoglobin glycosylation index (HGI), BMI, and age. The lowest risk group (with HGI < 0.44, BMI < 30 kg/m2, and age 0.44) experienced a 3.7% increase in absolute mortality risk attributable to intensive glycemic therapy. These mortality variations in the intensive treatment group were previously not detected by older, univariate subgroup analyses (6). Another study developed a prototype support vector regression model that outperformed diabetologists in predicting blood glucose levels at 30 and 60 min from a given time in patients with type 1 diabetes, and predicted about one quarter of hypoglycemic events 30 min ahead of the actual event (7). AI/ML-based algorithms have been extensively utilized and validated for diagnosis and classification of diabetic retinopathy (8–11). Deep learning systems and even purely database-driven AI algorithms have demonstrated the ability to diagnose diabetic retinopathy and related retinal diseases in large, multiethnic cohorts with high degree of sensitivity and specificity (8, 10). ML algorithms have also demonstrated the ability to incorporate associated risk factors such as duration of diabetes and insulin use into risk-stratification of diabetic retinopathy, which could potentially facilitate the development of better clinical decision support systems (11). A proprietary system IDx (Iowa City, IA) that uses ML technology to analyze retinal images in diabetic retinopathy had a sensitivity of 87% and specificity of 91% for autonomous detection of disease and received FDA approval in 2018 (12). This example represents the first prospective assessment of AI/ML in the clinic (5). AI/ML technologies has also been used in the analysis of large datasets generated from genomic technology. Deep-coverage whole-genome sequencing performed in 8,392 individuals of European and African descent to identify single-nucleotide variants and copy-number variations in Lipoprotein (a) revealed that LPA risk genotypes conferred greater relative risk for incident cardiovascular disease than direct measurements of Lipoprotein (a) levels. These risk genotypes were also associated with increased sub-clinical atherosclerotic disease in individuals of African ancestry (13). ML techniques were utilized in developing a novel mRNA based molecular test to detect BRAF V600E mutations in thyroid fine needle aspirate samples, which demonstrated sensitivity equal to that of established DNA-based assay and had lower non-diagnostic rates (14). By utilizing functional enrichment analysis followed by module analysis performed on protein-protein interaction network, the differential gene expression in anaplastic thyroid carcinoma was assessed (15). There were 247 up-regulated genes which were predominantly involved in cell cycle and 275 down-regulated genes that were mostly involved in thyroid hormone synthesis, insulin resistance, and cancer pathways, thus expanding on the...This publication has 28 references indexed in Scilit:
- Characteristics Associated With Decreased or Increased Mortality Risk From Glycemic Therapy Among Patients With Type 2 Diabetes and High Cardiovascular Risk: Machine Learning Analysis of the ACCORD TrialDiabetes Care, 2017
- Automatic Detection of Acromegaly From Facial Photographs Using Machine Learning MethodsEBioMedicine, 2017
- Development and Validation of a Deep Learning System for Diabetic Retinopathy and Related Eye Diseases Using Retinal Images From Multiethnic Populations With DiabetesJAMA, 2017
- Automated Identification of Diabetic Retinopathy Using Deep LearningOphthalmology, 2017
- Artificial Intelligence Methodologies and Their Application to DiabetesJournal of Diabetes Science and Technology, 2017
- Artificial intelligence in medicineMetabolism, 2017
- MALDI mass spectrometry imaging analysis of pituitary adenomas for near-real-time tumor delineationProceedings of the National Academy of Sciences of the United States of America, 2015
- MACHINE LEARNING FROM CONCEPT TO CLINIC: RELIABLE DETECTION OF BRAF V600E DNA MUTATIONS IN THYROID NODULES USING HIGH-DIMENSIONAL RNA EXPRESSION DATABiocomputing 2020, 2014
- Urine Steroid Metabolomics as a Biomarker Tool for Detecting Malignancy in Adrenal TumorsJournal of Clinical Endocrinology & Metabolism, 2011
- A hybrid machine learning-based method for classifying the Cushing's Syndrome with comorbid adrenocortical lesionsBMC Genomics, 2008