Identification of Important Symptoms and Diagnostic Hypothyroidism Patients Using Machine Learning Algorithms

Identification of Important Symptoms of Hypothyroidism, Prediction Hypothyroidism Patients Using Machine Learning Algorithms

Hypothyroidism, Machine learning algorithms, Random Forest, Decision tree, Logistic regression

Hypothyroidism (HT) is one of the most common endocrine diseases. It is, however, usually challenging for physicians to diagnose due to non-specific symptoms. The usual procedure for diagnosis of HT is a blood test. In recent years, machine learning algorithms have proved to be powerful tools in medicine due to their diagnostic accuracy. In this study, we aim to predict and identify the most important symptoms of HT using machine learning algorithms.

Hypothyroidism (HT) is one of the most common diseases in the world, in which insufficient thyroid hormone is produced. Due to the wide variation in clinical symptoms, the definition of HT is mainly biochemical. Ninety nine percent of primary cases of HT are related to deficiency of thyroxine (T4) and triiodothyronine (T3) hormones. Deficiency in T4 and T3 hormones, which are produced by thyroid gland, leads to increasing thyroid-stimulating hormone (TSH) production through a negative feedback mechanism . HT has non-specific symptoms such as weight gain, fatigue, insufficient concentration, depression, menstrual irregularities, and constipation, which change with age, gender, and other factors. Autoimmune thyroiditis (Hashimoto's disease) is the most common symptom of this disorder. The prevalence of HT is 2% in the world, even in the existence of enough iodine in daily food. In a cohort study that was conducted in Iran in 2017, a significant increase in the prevalence of thyroid dysfunction was reported, from 1.4 to 10.5, attributed to several factors such as geographical areas, aging, ethnicity and the amount of iodine intake. Increasing in serum cholesterol levels and the risk of coronary artery disease and cardiovascular mortality are the most common complications of HT. The economic burden of HT is fairly high, especially in patients with other underlying diseases such as diabetes and hemodialysis. The common clinical method for diagnosing equally primary HT is to check the serum concentration of TSH; People with TSH and T4 levels above the reference age range are diagnosed as hypothyroid. The upper limit of the TSH reference range usually increases with age in adults . In recent years, artificial intelligence and machine learning techniques have attracted increasing attention from medical researchers. Among the most attractive features of machine learning in medicine are disease prediction and diagnosis of simple symptoms . The prediction models such as support vector machine (SVM), decision tree (DT), random forest (RF) and artificial neural network (ANN), are among the most popular machine learning methods. As accurate diagnostic of HT is currently based on the TSH level obtained by a blood test, it creates some expense burden and anxiety for patients. The aim of the present study is to first diagnose HT in new cases that have no history of HT symptoms with three statistical machine learning methods (logistic regression, decision tree and random forest). The diagnosis is performed using simple and widely-accepted visual symptoms of HT that endocrinologists identify. Second, the most important visual features of HT which can help physicians in diagnosis, are also ranked using decision tree and random forest methods.

No related papers found yet.

Iran