Abstract
Background and aim
Artificial intelligence (AI) has emerged as a promising technology in the field of endocrinology, offering significant potential to revolutionize the diagnosis, treatment, and management of endocrine disorders. This comprehensive review aims to provide a concise overview of the current landscape of AI applications in endocrinology and metabolism, focusing on the fundamental concepts of AI, including machine learning algorithms and deep learning models.
Methods
The review explores various areas of endocrinology where AI has demonstrated its value, encompassing screening and diagnosis, risk prediction, translational research, and “pre-emptive medicine”. Within each domain, relevant studies are discussed, offering insights into the methodology and main findings of AI in the treatment of different pathologies, such as diabetes mellitus and related disorders, thyroid disorders, adrenal tumors, and bone and mineral disorders.
Results
Collectively, these studies show the valuable contributions of AI in optimizing healthcare outcomes and unveiling new understandings of the intricate mechanisms underlying endocrine disorders. Furthermore, AI-driven approaches facilitate the development of precision medicine strategies, enabling tailored interventions for patients based on their individual characteristics and needs.
Conclusions
By embracing AI in endocrinology, a future can be envisioned where medical professionals and AI systems synergistically collaborate, ultimately enhancing the lives of individuals affected by endocrine disorders.
Similar content being viewed by others
Explore related subjects
Find the latest articles, discoveries, and news in related topics.Avoid common mistakes on your manuscript.
Introduction
The exponential growth of technology in the past two decades has paved for the development of advanced techniques capable of addressing scientific inquiries at a magnitude far surpassing human capabilities. One notable example is the field of artificial intelligence (AI). AI is a branch of computer science that focuses on the theory and development of computer systems and algorithms capable of performing tasks that typically require human intelligence [1]. The healthcare sector is currently undergoing an unprecedented transformation due to AI, as it possesses the potential to enhance existing clinical practices. The innovative aspects introduced by AI find ideal applications within the field of endocrinology, given its complex and interconnected nature. Indeed, unlike other medical domains, endocrinology is not related to a single organ structure but is a complicated biological system of hormones and metabolites, where hormones function within an elaborate network of local and remote actions involving receptors, signaling pathways, and intricate feedback mechanisms [2, 3]. These complex and interconnected systems are often beyond the comprehension and reasoning abilities of the human brain. The purpose of this study is to explore the diverse applications of AI in the field of endocrinology and metabolism, with a focus on its potential to enhance screening, disease diagnosis, risk prediction, prognosis, and medical research. Consequently, it offers an overview of AI's capacity to gather valuable information, deliver personalized care, and enhance patient outcomes within the dynamic field of endocrine and metabolic diseases.
Artificial intelligence: a brief introduction
The terms artificial intelligence, machine learning (ML), and deep learning (DL) are often used concurrently and sometimes interchangeably in medical literature. However, they are not synonyms. AI is the field that encompasses the theory and advancement of computer systems and algorithms designed to execute tasks that commonly necessitate human intelligence; ML is a subset of AI techniques, including those methods that use statistical learning to allow machines to improve with experience. DL, in turn, is a subset of machine learning techniques that use complex algorithms inspired by the human brain and how it works [4]. In the medical field, the class of algorithms that is being developed and used the most is ML. ML can be classified into four main categories: supervised, unsupervised, semi-supervised, and reinforcement learning as presented in Table 1 [5].
Research investigations in the medical field related to AI, particularly based on ML, encompass a complex workflow. This workflow comprises problem formulation, data acquisition and preprocessing, the selection of a suitable ML model, and culminates in the comprehensive evaluation of the model (Fig. 1). It is crucial to emphasize that this workflow serves as a general outline and acknowledges that specific steps and details may vary depending on the research problem, available resources, and the requirements of the medical domain under study. For more comprehensive insights, refer to [5, 6].
Methods
The comprehensive review we propose aims to achieve a concise overview of the landscape of AI applications in endocrinology and metabolism, with the goal of demonstrating its capability to collect valuable information, offer personalized care, and enhance patient outcomes. To achieve this objective, we employed a targeted search and selection strategy, which will be outlined in the following sections.
Search strategy
F.G. conducted a comprehensive literature review by querying the PubMed database. Any disagreements were resolved by the senior reviewers S.D. and G.D.D. until an agreement was reached on all issues. Our search strategy involved a combination of specific keywords, including “machine learning”, “deep learning”, “pre-emptive medicine”, “diagnosis”, “prediction”, “risk prediction”, “detection” and “treatment” as well as Medical Subject Headings (MeSH) terms such as “diabetes”, “diabetes disorders”, “adrenal tumors”, “thyroid nodules”, and “bone and mineral disorders”. We connected these keywords and MeSH terms using “AND”, such as “machine learning” AND “diagnosis” AND “diabetes”, “machine learning” AND “prediction” AND “diabetes”, “machine learning” AND “detection” AND “diabetes disorders”, “pre-emptive medicine” AND “diabetes disorders” and so on. All conducted research was restricted to studies published from 2017 to the present, with a focus on articles available in full-text English. In this procedure, we also considered grey-literature sources, notably integrating selected proceedings [7, 8] and a report from a research institute [9] with the aim of elucidating the characteristics of specific diseases and highlighting the innovative aspects of “pre-emptive medicine”.
Study selection
The study selection process was carried out in two stages: an initial screening of titles and abstracts, followed by a more detailed assessment of full-text articles. Eligible studies were included if they met the following three criteria: (1) ensuring an adequate number of subjects (n > 10) representing the variety of disease categories, including diabetes and related disorders, adrenal tumors, thyroid disorders, and bone and mineral disorders; (2) encompassing studies on ML and DL, with at least one study illustrating different ML applications such as supervised, unsupervised, and reinforcement learning; and (3) prioritizing studies with recent publication dates whenever possible. Subsequently, we classified these studies according to the domains in which AI plays a significant role in the field of endocrinology and metabolism. These domains include screening and diagnosis, risk prediction, translational research, and the emerging field of “pre-emptive medicine”, which may exhibit overlapping elements in certain cases. Furthermore, it is important to acknowledge that while these selected studies may not fully capture the entire spectrum of AI applications in the field of endocrinology, they provide practical examples that help illustrate the utility of ML and DL algorithms across various areas of endocrine research.
Results
Among the 319 articles that we initially identified, 168 were removed after evaluating the titles and removing duplicates. Among the remaining 151 studies, 115 were excluded after reviewing their abstracts, since they did not satisfy the inclusion criteria. We conducted a meticulous examination of the remaining 36 full-text articles. Ultimately, 17 articles [10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26] met the predefined inclusion criteria. The flowchart of the included studies is presented in Fig. 2 while the main features of the included studies are summarized in Table 2.
Screening and diagnosis
AI has revolutionized the field of screening and diagnosis, significantly improving the accuracy, efficiency, and effectiveness of medical assessments. In particular, AI aims to enhance screening strategies, given their significant clinical impact on endocrine disorders. Additionally, it aims to streamline the diagnostic workflow by analyzing extensive patient data, including medical records, imaging scans, and laboratory results, enabling faster, more precise, and efficient evaluations. Furthermore, AI is poised to discover novel disease clusters and associations by identifying previously unknown patterns and connections within complex medical data, thereby expanding our understanding of diseases themselves. To demonstrate the tangible impact brought about by AI in screening and diagnosis, we will present several applications and categorize them based on the specific objectives AI seeks to achieve.
Improvement of screening strategies
Efficient screening tools for endocrine disorders have the potential to bring about significant clinical benefits. These benefits include improving the prognosis of individual patients by enabling earlier detection of diseases, as well as optimizing the allocation of public health resources through targeted focus on high-risk individuals and avoiding unnecessary testing in low-risk groups. In this context, researchers have been exploring the capabilities of ML and DL algorithms to determine whether they can offer a superior approach to screening for different endocrine diseases. For instance, multiple studies have demonstrated the application of AI in the field of diabetes and related disorders, illustrating its potential for early diagnosis and the development of therapeutic strategies aimed at preventing or postponing the onset of complications. Agliata et al. set out to use a supervised ML technique to investigate the correlations between an individual's health status and the development of type 2 diabetes, aiming to accurately predict its onset or assess the individual's risk level [10]. They proposed the implementation of a binary classifier with a shallow architecture, specifically a neural network, trained from scratch, to detect potential non-linear associations between the onset of type 2 diabetes and a collection of parameters derived from patient measurements. The conducted ablation study by the researchers revealed that the binary classifier, optimized using the Adam algorithm, achieved a satisfactory level of accuracy (approximately 86% on the test set) and a receiver operating characteristic (ROC) area under the curve (AUC) value of 0.934 in predicting the onset of diabetes in non-linear relationships with specific patient measurements. This neural network-based approach holds promise in delivering accurate information for personalized medicine, thereby serving as a valuable resource for decision-making. Furthermore, AI-based algorithms have been extensively utilized and validated for the diagnosis and classification of diabetic retinopathy (DR). Raju et al. proposed a DL model consisting of a convolutional neural network (CNN) based approach to classify the stage of diabetic retinopathy and detect the laterality of the eye using funduscopic images [11]. They utilized the publicly accessible Kaggle DR dataset composed of 88,702 color fundus images, including 35,126 samples for training and 53,576 samples for testing, exhibiting a sensitivity of 80.28% and a specificity of 92.29% on the test set. Furthermore, the network was trained on 8810 images to detect the laterality of the eye, achieving an accuracy of 93.28% on the validation set consisting of 8816 images [11]. AI is also widely used in the field of distinguishing and differentiating adrenal tumors through the utilization of imaging techniques such as computed tomography (CT). Liu et al. proposed the application of supervised ML prediction models and scoring systems to differentiate between subclinical pheochromocytoma (sPHEO) and lipid-poor adenoma (LPA) [12]. Specifically, they employed logistic regression (LR), support vector machine (SVM), and random forest (RF) approaches to assess the accuracy of CT-based ML models in distinguishing sPHEO from LPA in patients with adrenal incidentalomas. The results demonstrated that the LR model outperformed the other models, achieving an AUC of 0.917 and an accuracy of 0.864. Medical image data holds the potential to offer relevant features that are well-suited for opportunistic screening of endocrine disorders. Valentinitsch et al. developed and trained a supervised ML model to detect prevalent vertebral fractures using non-fractured vertebral regions from CT scans performed for various reasons [13]. By incorporating global and local density and texture parameters, the ML model exhibited superior performance compared to relying solely on volumetric bone mineral density (BMD) in discerning the presence of vertebral fractures. These findings suggest the potential of a semi-automated pipeline for opportunistically screening individuals at high risk of fractures.
Facilitation of the diagnostic workflow
The facilitation of the diagnostic workflow is a crucial element in modern healthcare, as timely and accurate diagnoses play a vital role in effective treatment and patient care. AI systems contribute to streamlining the diagnostic process by analyzing large volumes of patient data, thereby assisting the decision-making process and reducing diagnostic uncertainty. Peng et al. developed a DL model named ThyNet to aid in the diagnosis and management of thyroid nodules [15]. ThyNet utilized ultrasound image sets to differentiate between malignant and benign tumors, enabling a strategy for clinical decision-making. The results of this study showed advantages in improving the diagnostic accuracy of radiologists on thyroid nodule differentiation and could potentially decrease the number of unnecessary fine needle aspirations. In certain diseases, a well-validated and accurate non-invasive ML or DL model may have the potential to replace standard invasive diagnostic methods. For instance, the global prevalence of nonalcoholic fatty liver disease (NAFLD) is experiencing a rapid increase. However, invasive liver biopsy continues to be the gold-standard method for diagnosing both NAFLD and nonalcoholic steatohepatitis. In a study by Perakakis et al., a supervised method consisting of an SVM model was developed to classify NAFLD [16]. This model utilized features obtained from lipidomic, glycomic and liver fatty acid analysis of serum samples. To detect liver fibrosis, a concise exploratory model focused on ten lipid species achieved high accuracy (up to 98%). This suggests that a targeted lipidomic approach holds promise as a non-invasive alternative diagnostic tool. However, it is essential to validate the model further across diverse ethnicities and individuals with varying degrees of liver disease severity. AI holds significant promise in mitigating diagnostic uncertainty, particularly in challenging domains such as asymptomatic hyperparathyroidism. The identification of this condition proves challenging due to its subtle biochemical alterations and overlapping phenotype with primary osteoporosis and other rare mineral disorders [7]. In a study by Somnay et al., a supervised ML model, specifically a Bayesian network, was trained to recognize primary hyperparathyroidism among patients who underwent neck surgery, such as thyroidectomy or parathyroidectomy [14]. However, the model exhibited relatively low performance in detecting mild disease.
Finding novel disease clusters and associations
Exploring novel disease clusters and associations offers valuable insights into the intricate network of biological pathways and interactions within the human body. In this regard, AI algorithms play a crucial role by analyzing extensive datasets from various sources, thereby revealing previously undiscovered connections between diseases that may have evaded traditional analytical approaches. These advanced algorithms have the ability to identify complex interactions, genetic variations, and environmental factors that contribute to the development and progression of diseases. Cho et al. proposed a method to identify distinct population clusters that exhibit variations in the development of type 2 diabetes [17]. At first, they employed a risk-factor-based clustering (RFC) approach, which involved hierarchically clustering the population using profiles of five established risk factors for type 2 diabetes: age, gender, body mass index, hypertension, and family history of diabetes. The RFC analysis successfully identified six population clusters in the discovery data, showing significantly different prevalence rates of type 2 diabetes within each cluster. After identifying the clusters in the discovery data, an SVM model was applied to validate the findings. The SVM model also identified six clusters in the validation data, further confirming the heterogeneity of type 2 diabetes prevalence across these clusters. Notably, beyond variations in diabetes prevalence, the identified clusters exhibited distinct clinical features, including variations in biochemical profiles, and demonstrated different prediction performances using the risk factors [17]. Furthermore, in this context, unsupervised learning can be a valuable tool for discovering novel clusters and associations within a given dataset. For instance, Marquardt et al. proposed the utilization of an unsupervised ML-based method to cluster adrenocortical tumors solely based on messenger ribonucleic acid (mRNA) expression [18]. Specifically, they employed a visual-based clustering method on the ribonucleic acid (RNA) sequencing data from a large cohort of adrenocortical carcinoma (ACC) patients obtained from The Cancer Genome Atlas (TCGA). This approach successfully classified the tumors into two distinct clusters, which were found to be correlated with patient survival outcomes. Applying the visual clustering method to a second dataset that included benign adrenocortical samples, the study further revealed that one of the ACC clusters exhibited closer proximity to the benign samples. This observation provided a potential explanation for the improved survival observed in this particular ACC cluster. Moreover, by employing ML techniques, the researchers identified novel potential biomarker genes with prognostic value for this rare disease. These genes exhibited significant differential expression across the distinct survival clusters and warrant further evaluation [18].
Risk prediction
AI algorithms can analyze extensive datasets encompassing patient information, laboratory results, imaging data, and genetic profiles to generate accurate risk prediction models for various endocrine disorders. By considering many factors and their complex interactions, AI can identify individuals at higher risk of developing conditions such as diabetes, thyroid disorders, or adrenal diseases. Furthermore, AI-driven models can evaluate treatment responses and predict patient outcomes based on clinical data, lifestyle factors, and treatment protocols. These predictive models enable endocrinologists to tailor treatment plans, optimize medication dosages, and make informed decisions regarding therapeutic interventions. In order to show the concrete advancements facilitated by AI in risk prediction, the subsequent applications are presented and classified into two categories: assessment of clinical outcomes and assessment of treatment responses.
Assessment of clinical outcomes
The ability to accurately predict clinical outcomes empowers healthcare professionals to adopt an individualized approach to treatment strategy and monitoring. Several investigations have been conducted, for example, to address this objective, focusing on diabetes complications (DCs) through the utilization of ML techniques. These techniques provide an opportunity to identify patients who are at a higher risk of experiencing complications. A study conducted by Nicolucci et al. [19] serves as a notable example in this field. This study has focused on six categories of DCs: eye complications, cardiovascular diseases, cerebrovascular diseases, peripheral vascular diseases, nephropathy, and diabetic neuropathy. They developed a supervised learning approach utilizing tree-based algorithms (XGBoost) to predict the occurrence of each complication within a span of 5 years (task 1), as well as separate predictions for early (within 2 years) and late (3–5 years) onset of complications (task 2). The results for all DCs demonstrated predictive models with an accuracy exceeding 70% and an AUC surpassing 0.80, reaching 0.97 for nephropathy in task 1. For task 2, all predictive models exhibited an accuracy above 70% and an AUC greater than 0.85. The sensitivity in predicting the early occurrence of complications ranged from 83.2% for peripheral vascular disease to 88.5% for nephropathy [19]. An additional example is illustrated in the detection of coronary artery atherosclerosis in individuals with type 2 diabetes mellitus [20]. This is achieved through the utilization of an unsupervised clustering analysis based on clinical factors, which aims to differentiate the population heterogeneity of type 2 diabetes and evaluate the differences in coronary atherosclerosis as evaluated through coronary computed tomography angiography (CCTA). This method exemplifies the capability to effectively address patients with heterogeneous clinical indicators and identify groups with different types of coronary plaque and degrees of coronary stenosis [20].
Assessment of treatment responses
ML and DL principles can be applied to predict treatment responses among patients affected by the same pathology. Teh et al. introduced a novel approach employing DL to predict the treatment response in individuals suffering from painful diabetic peripheral neuropathy (pDPN) [22]. They used resting-state functional magnetic resonance imaging (rs-fMRI) to extract functional connectivity features by means of group independent component analysis (gICA). Subsequently, they developed an automated treatment response classification model using three-dimensional convolutional neural networks (3D-CNN) to effectively distinguish between responders and non-responders to lidocaine treatment, showing the potential of deep learning in accurately predicting treatment outcomes for pDPN patients. Moreover, efficient ML and DL models hold promising potential in offering guidance for dose adjustment, particularly among patients with chronic conditions. For example, a reinforcement learning (RL) algorithm was developed to aid in determining the optimal dosage of long-acting insulin for individuals diagnosed with type 1 diabetes, utilizing clinical data [21]. This study demonstrates that an RL algorithm can be employed to provide personalized insulin doses, ensuring sufficient glycemic control in patients with type 1 diabetes. However, further investigation involving a larger patient sample is required to validate these findings. Another compelling example is provided by Zaborek et al., who constructed a supervised ML model to facilitate levothyroxine dose adjustment following thyroidectomy [23]. Their findings revealed a notable enhancement in predictive accuracy compared to the prevailing weight-based dosing approach, thereby demonstrating a substantial improvement.
Translational research
ML algorithms have become a crucial methodology in translational research with the rise of the multi-omics approach, which produces abundant datasets with numerous features to be accounted for. Liu et al. developed an ML algorithm that integrated baseline microbial signatures to identify crucial microbiota species and metabolites strongly associated with exercise responsiveness in humans [24]. They observed distinct patterns of exercise-induced alterations in the gut microbiota between human exercise responders and non-responders. Moreover, through fecal microbial transplantation from responders to mice, they demonstrated that the benefits of exercise on insulin sensitivity could be conferred. By employing a random forest algorithm, they selected 19 features, among species and metabolites, which exhibited significant differences between the exercise-responsive and non-responsive groups. These selected features, among the numerous microbiota species and metabolites investigated, hold potential as biomarkers for personalized responses to exercise [24]. Another study aimed to identify proteomics-based biomarkers associated with various health outcomes, such as percentage body fat, lean mass, current smoking, and the risk of developing cardiovascular complications [25]. To accomplish this, the researchers adopted a comprehensive approach by leveraging extensive community-based cohort databases and samples. Employing ML techniques, they successfully discovered a set of highly predictive proteins and developed corresponding models. However, it is important to note that the practical application and generalizability of these findings must be confirmed through long-term studies conducted in diverse populations [25].
Pre-emptive medicine
"Pre-emptive medicine" is an emerging field that leverages AI technology with the potential for extensive future applications. Originating in Japan, this concept aims to accurately anticipate the onset and progression of diseases by utilizing genomic information, biomarkers, bioimages, and other biological data. Its goal is to provide therapeutic interventions for diseases at their early stages, even before symptoms manifest in individuals. The concept of pre-emptive medicine takes into account the time-course of a disease in each individual and strives to employ medical interventions to prevent disease progression. Non-communicable diseases such as hypertension or diabetes [9] are particularly suitable and promising targets for pre-emptive medicine [8, 26]. For instance, in the context of pre-emptive medicine in hypertension, the ultimate goal is to completely prevent the onset of the pathology by precisely predicting the elevation of blood pressure, even in individuals with normal blood pressure or at early stages of hypertension [26]. To accomplish this, it is crucial to detect abnormal fluctuations in blood pressure as the earliest manifestation of the disease in an individual. Using the DL method, this analysis identifies changes in various biological data points that lead to increases or variations in blood pressure. By examining the chronologically accumulated biological data, it can also predict the future course of hypertension in an individual.
AI limitations in medicine
AI algorithms heavily rely on the quality and quantity of data they are trained on [27]. Inaccurate or biased data can lead to flawed predictions and diagnoses, potentially compromising patient safety and outcomes [28]. Moreover, the issue of data privacy and patient confidentiality remains a significant concern, as the utilization of sensitive medical data for AI training purposes must adhere to stringent ethical and regulatory standards [29,30,31]. Another limitation arises from the "black-box" nature of some AI models, particularly in DL [27, 32]. Understanding how these models arrive at specific decisions can be challenging, hindering their acceptance among medical professionals who require transparency and interpretability in clinical decision-making [33]. Additionally, the integration of AI tools into existing healthcare systems and workflows poses practical challenges, including compatibility issues, staff training, and the need for substantial financial investments [33]. Furthermore, the regulatory landscape surrounding AI in medicine is continually evolving, and navigating these regulations while ensuring patient safety and efficacy can be a complex endeavor [34]. AI systems must meet stringent validation and verification criteria before widespread adoption can occur. Lastly, while AI algorithms can significantly enhance clinical decision-making, they should always complement rather than replace human expertise [35, 36]. Indeed, maintaining a human-in-the-loop approach cannot be overstated, as medical professionals remain essential for contextual understanding and ethical decision-making.
Conclusions
In conclusion, this study has explored the remarkable potential of AI in the field of endocrinology by providing diverse examples of its applications. Through advancements in ML and DL, AI has demonstrated its ability to enhance various aspects of endocrine research and clinical practice. Improved screening, disease diagnosis, risk prediction, personalized treatment, and patient management are among the valuable contributions AI offers for optimizing healthcare outcomes in endocrinology. The application of AI algorithms in analyzing complex data sets has opened up new ways for understanding the intricate mechanisms underlying endocrine disorders. Moreover, AI-driven approaches enable the development of precision medicine strategies, offering tailored interventions for patients based on their individual characteristics and needs. As AI continues to evolve, it holds immense promise for transforming endocrinology by enabling more accurate diagnoses, potentially reducing unnecessary investigations, improving patient outcomes, reducing healthcare expenditures, facilitating efficient digital storage of vast patient data, and contributing to advancements in our understanding and management of endocrine-related diseases. Embracing AI in endocrinology can lead to a future where medical professionals and AI systems work synergistically, ultimately improving the lives of individuals affected by endocrine disorders.
Data availability
Not applicable.
Code availability
Not applicable.
References
Parag Marajan M (2022) Artificial intelligence in healthcare, 2nd (general) edn. Dr Parag Suresh Mahajan, London
Ashwell E (2022) The endocrine system and associated disorders. Br J Nurs 31:316–320. https://doi.org/10.12968/bjon.2022.31.6.316
Oikonomakos IT, Steenblock C, Bornstein SR (2023) Artificial intelligence in diabetes mellitus and endocrine diseases—What can we expect? Nat Rev Endocrinol 19:375–376. https://doi.org/10.1038/s41574-023-00852-1
Alpaydin E (2009) Introduction to machine learning, 2nd edn. The MIT Press, London
Hamet P, Tremblay J (2017) Artificial intelligence in medicine. Metabolism 69S:S36-40. https://doi.org/10.1016/j.metabol.2017.01.011
Amisha, Malik P, Pathania M, Rathaur VK (2019) Overview of artificial intelligence in medicine. J Family Med Prim Care 8:2328–2331. https://doi.org/10.4103/jfmpc.jfmpc_440_19
Eastell R, Brandi ML, Costa AG, D’Amour P, Shoback DM, Thakker RV (2014) Diagnosis of asymptomatic primary hyperparathyroidism: proceedings of the Fourth International Workshop. J Clin Endocrinol Metab 99:3570–3579. https://doi.org/10.1210/jc.2014-1414
Imura H (2013) Life course health care and preemptive approach to non-communicable diseases. Proc Jpn Acad Ser B 89:462–473. https://doi.org/10.2183/pjab.89.462
Kato T, Kinoshita M (2017) Preemptive medicine—a case study of diabetes. Mitsui & Co. Global Strategic Studies Institute Monthly Report
Agliata A, Giordano D, Bardozzo F, Bottiglieri S, Facchiano A, Tagliaferri R (2023) Machine learning as a support for the diagnosis of type 2 diabetes. Int J Mol Sci 24:6775. https://doi.org/10.3390/ijms24076775
Raju M, Pagidimarri V, Barreto R, Kadam A, Kasivajjala V, Aswath A (2017) Development of a deep learning algorithm for automatic diagnosis of diabetic retinopathy. Stud Health Technol Inform 245:559–563
Liu H, Guan X, Xu B, Zeng F, Chen C, Yin H et al (2022) Computed tomography-based machine learning differentiates adrenal pheochromocytoma from lipid-poor adenoma. Front Endocrinol (Lausanne) 13:833413. https://doi.org/10.3389/fendo.2022.833413
Valentinitsch A, Trebeschi S, Kaesmacher J, Lorenz C, Löffler MT, Zimmer C et al (2019) Opportunistic osteoporosis screening in multi-detector CT images via local classification of textures. Osteoporos Int 30:1275–1285. https://doi.org/10.1007/s00198-019-04910-1
Somnay YR, Craven M, McCoy KL, Carty SE, Wang TS, Greenberg CC et al (2017) Improving diagnostic recognition of primary hyperparathyroidism with machine learning. Surgery 161:1113–1121. https://doi.org/10.1016/j.surg.2016.09.044
Peng S, Liu Y, Lv W, Liu L, Zhou Q, Yang H et al (2021) Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and management: a multicentre diagnostic study. Lancet Digital Health 3:e250–e259. https://doi.org/10.1016/S2589-7500(21)00041-8
Perakakis N, Polyzos SA, Yazdani A, Sala-Vila A, Kountouras J, Anastasilakis AD et al (2019) Non-invasive diagnosis of non-alcoholic steatohepatitis and fibrosis with the use of omics and supervised learning: a proof of concept study. Metabolism 101:154005. https://doi.org/10.1016/j.metabol.2019.154005
Cho SB, Kim SC, Chung MG (2019) Identification of novel population clusters with different susceptibilities to type 2 diabetes and their impact on the prediction of diabetes. Sci Rep 9:3329. https://doi.org/10.1038/s41598-019-40058-y
Marquardt A, Landwehr L-S, Ronchi CL, Di Dalmazi G, Riester A, Kollmannsberger P et al (2021) Identifying new potential biomarkers in adrenocortical tumors based on mRNA expression data using machine learning. Cancers 13:4671. https://doi.org/10.3390/cancers13184671
Nicolucci A, Romeo L, Bernardini M, Vespasiani M, Rossi MC, Petrelli M et al (2022) Prediction of complications of type 2 diabetes: a machine learning approach. Diabetes Res Clin Pract. https://doi.org/10.1016/j.diabres.2022.110013
Jiang Y, Yang Z-G, Wang J, Shi R, Han P-L, Qian W-L et al (2022) Unsupervised machine learning based on clinical factors for the detection of coronary artery atherosclerosis in type 2 diabetes mellitus. Cardiovasc Diabetol 21:259. https://doi.org/10.1186/s12933-022-01700-8
Oroojeni Mohammad Javad M, Agboola SO, Jethwani K, Zeid A, Kamarthi S (2019) A reinforcement learning-based method for management of type 1 diabetes: exploratory study. JMIR Diabetes 4:e12905. https://doi.org/10.2196/12905
Teh K, Armitage P, Tesfaye S, Selvarajah D (2023) Deep learning classification of treatment response in diabetic painful neuropathy: a combined machine learning and magnetic resonance neuroimaging methodological study. Neuroinform 21:35–43. https://doi.org/10.1007/s12021-022-09603-5
Zaborek NA, Cheng A, Imbus JR, Long KL, Pitt SC, Sippel RS et al (2019) The optimal dosing scheme for levothyroxine after thyroidectomy: a comprehensive comparison and evaluation. Surgery 165:92–98. https://doi.org/10.1016/j.surg.2018.04.097
Liu Y, Wang Y, Ni Y, Cheung CKY, Lam KSL, Wang Y et al (2020) Gut microbiome fermentation determines the efficacy of exercise for diabetes prevention. Cell Metab 31:77-91.e5. https://doi.org/10.1016/j.cmet.2019.11.001
Williams SA, Kivimaki M, Langenberg C, Hingorani AD, Casas JP, Bouchard C et al (2019) Plasma protein patterns as comprehensive indicators of health. Nat Med 25:1851–1857. https://doi.org/10.1038/s41591-019-0665-2
Itoh H, Hayashi K, Miyashita K (2019) Pre-emptive medicine for hypertension and its prospects. Hypertens Res 42:301–305. https://doi.org/10.1038/s41440-018-0177-3
Khan B, Fatima H, Qureshi A, Kumar S, Hanan A, Hussain J et al (2023) Drawbacks of artificial intelligence and their potential solutions in the healthcare sector. Biomed Mater Dev. https://doi.org/10.1007/s44174-023-00063-2
Rajkomar A, Dean J, Kohane I (2019) Machine learning in medicine. N Engl J Med 380:1347–1358. https://doi.org/10.1056/NEJMra1814259
Price WN, Cohen IG (2019) Privacy in the age of medical big data. Nat Med 25:37–43. https://doi.org/10.1038/s41591-018-0272-7
Rajpurkar P, Chen E, Banerjee O, Topol EJ (2022) AI in health and medicine. Nat Med 28:31–38. https://doi.org/10.1038/s41591-021-01614-0
Hatherley JJ (2020) Limits of trust in medical AI. J Med Ethics 46:478–481. https://doi.org/10.1136/medethics-2019-105935
Rudin C (2019) Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat Mach Intell 1:206–215. https://doi.org/10.1038/s42256-019-0048-x
Price WN, Gerke S, Cohen IG (2019) Potential liability for physicians using artificial intelligence. JAMA 322:1765–1766. https://doi.org/10.1001/jama.2019.15064
Wang F, Casalino LP, Khullar D (2019) Deep learning in medicine-promise, progress, and challenges. JAMA Intern Med 179:293–294. https://doi.org/10.1001/jamainternmed.2018.7117
Topol EJ (2019) High-performance medicine: the convergence of human and artificial intelligence. Nat Med 25:44–56. https://doi.org/10.1038/s41591-018-0300-7
Sezgin E (2023) Artificial intelligence in healthcare: complementing, not replacing, doctors and healthcare providers. Digit Health 9:20552076231186520. https://doi.org/10.1177/20552076231186520
Funding
Open access funding provided by Alma Mater Studiorum - Università di Bologna within the CRUI-CARE Agreement. The research leading to these results has received funding from the European Union—NextGenerationEU through the Italian Ministry of University and Research under PNRR—M4C2-I1.3 Project PE_00000019 "HEAL ITALIA" to Stefano Diciotti" CUP J33C22002920006. The views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Commission. Neither the European Union nor the European Commission can be held responsible for them.
Author information
Authors and Affiliations
Contributions
SD proposed the idea, revised and edited the manuscript. FG designed the review protocol, conducted the search and screened potentially eligible studies. GDD critically revised the manuscript. SD and GDD resolved any disagreements about the inclusion of some documents until an agreement was reached on all issues. All authors contributed to and agreed on the final version of the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Consent to participate
Not applicable
Consent for publication
Not applicable.
Ethical approval
This study dealt only with data used in the citated articles, no ethical approval was needed.
Informed consent
For this type of study, consent is not required.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Giorgini, F., Di Dalmazi, G. & Diciotti, S. Artificial intelligence in endocrinology: a comprehensive review. J Endocrinol Invest 47, 1067–1082 (2024). https://doi.org/10.1007/s40618-023-02235-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40618-023-02235-9