Natural occurrence of nocturnal hypoglycemia detection using hybrid particle swarm optimized fuzzy reasoning model

Abstract

Introduction

Low blood glucose (hypoglycemia) is a common and serious side effect of insulin therapy in patients with diabetes. This paper will make a contribution to knowledge in the modeling and design of a non-invasive hypoglycemia monitor for patients with type 1 diabetes mellitus (T1DM) using a fuzzy-reasoning system.

Methods

Based on the heart rate and the corrected QT interval of the electrocardiogram (ECG) signal, we have developed a hybrid particle-swarm-optimization-based fuzzy-reasoning model to recognize the presence of hypoglycemic episodes. To optimize the fuzzy rules and the fuzzy-membership functions, a hybrid particle-swarm-optimization with wavelet mutation operation is investigated.

Results

From our clinical study of 16 children with T1DM, natural occurrence of nocturnal-hypoglycemic episodes was associated with increased heart rates and increased corrected QT intervals. All the data sets were collected from the Government of Western Australia's Department of Health. All data were organized randomly into a training set (8 patients with 320 data points) and a testing set (another 8 patients with 269 data points). To prevent the phenomenon of overtraining, we separated the training set into 2 sets (4 patients in each set) and a fitness function was introduced for this training process. The testing performances of the proposed algorithm for detection of advanced hypoglycemic episodes (sensitivity = 85.71% and specificity = 79.84%) and hypoglycemic episodes (sensitivity = 80.00% and specificity = 55.14%) were given.

Conclusion

We have investigated the detection for the natural occurrence of nocturnal hypoglycemic episodes in T1DM using a hybrid particle-swarm-optimization-based fuzzy-reasoning model with physiological parameters. In this study, no restricted environment (e.g. patient's dietary requirements) is required. Furthermore, the sampling time is between 5 and 10 min. To conclude, we have shown that the testing performances of the proposed algorithm for detection of advanced hypoglycemic and hypoglycemic episodes for T1DM patients are satisfactory.

Introduction

In Australia, 1.7 million people suffer from diabetes [1]. Many of these will experience grave medical conditions, such as heart and kidney failure, amputation and blindness and some will die as a consequence of a low blood-glucose level (hypoglycemia). Hypoglycemia is the most common complication experienced by patients with insulin-dependent diabetes mellitus. Its onset is characterized by sweating, tremor, palpitations, loss of concentration and tiredness [2]. Although many patients can recognize the onset of hypoglycemia and take corrective action, a significant number of patients are unable to recognize the onset of symptoms, with life-threatening consequences.

Hypoglycemia in diabetic patients has the potential to become dangerous. In general, the blood-glucose level in men can drop to 3 mmol/l (54 mg/dl) after 24 h of fasting and to 2.7 mmol/l after 72 h of fasting. In women, the blood-glucose level can be as low as 2 mmol/l after 24 h of fasting [3]. Blood-glucose levels below 2.5 mmol/l are almost always associated with serious complications. In many cases of hypoglycemia, the symptoms can occur without the patient's knowledge [4] and at any time, e.g. while driving or during sleep. In severe cases, the patient can lapse into a coma and die. Nocturnal episodes are potentially dangerous and have been implicated when diabetic patients have died unexpectedly in their sleep. Hypoglycemia is one of the complications of diabetes most feared by patients, on a par with blindness and renal failure [5].

In this paper, blood glucose levels (BGL) <3.33 mmol/l (60 mg/dl) are considered as hypoglycemic episodes and BGL < 2.80 mmol/l (50 mg/dl) as advanced hypoglycemic episodes [3]. Nocturnal hypoglycemia is particularly dangerous because sleep reduces and may obscure autonomic counter-regulatory responses, so that an initially mild episode may become severe. The risk of severe hypoglycemia is high at night, since 50% of all severe episodes occur at that time [6]. Even with modest insulin elevations, deficient glucose counter-regulation may also lead to severe hypoglycemia. Regulation of nocturnal hypoglycemia is further complicated by the dawn phenomenon. This is a consequence of nocturnal changes in insulin sensitivity secondary to growth hormone secretion: a decrease in insulin requirements approximately between midnight and 5 am followed by an increase in requirements between 5 am and 8 am.

This paper sheds light on the modeling and design of a non-invasive and small hypoglycemia monitor, using physiological responses and a fuzzy-reasoning system; patients with type 1 diabetes mellitus (T1DM) can use it anywhere, e.g. in the home, hospital or office. There are some non-invasive blood-glucose monitoring systems currently available but each has specific drawbacks in terms of reliability and obtrusiveness. Intensive research has been devoted to the development of hypoglycemia alarms, exploiting principles that range from detecting changes in skin conductance (due to sweating) to measurements, by glucose sensors, of subcutaneous tissue glucose concentrations [6]. However, none of these has proved sufficiently reliable or unobtrusive.

Although real-time continuous-glucose-monitoring systems (CGMS) are now available to give real-time estimations of glucose levels, these still have some drawbacks when used as an alarm. For the MiniMed Medtronic (Northridge, CA) CGMS, the median error was reported as 10–15% at a plasma glucose of 4–10 mmol/l [7]. The efficacy of CGMS in detecting unrecognized hypoglycemia has been performed with 79.1% sensitivity and 97.5% specificity [8]. The limitation of this study was that patients were asked not to change their dietary regimens. For the Abbott freestyle navigator CGMS, the sensitivity and specificity during hypoglycemia (3.9 mmol/l or 70 mg/dl) were reported as 79.8% and 92.8% respectively [9]. However, the sampling time is reported within ±30 min, which is too long for an instant alarm. It means that a hypoglycemia episode within 30 min was considered a single hypo event and counted only once in the evaluation of sensitivity. For the Solianis’ multisensory system, the median-absolute relative difference is 27.3% for testing data [10]. As these are median values, the errors may be significantly greater and, as a result, the manufacturers do not recommend its use as an alarm. In [11], the detection of nocturnal hypoglycemic episodes in children with type 1 diabetes using Bayesian neural network is presented. Primary results are given to show the correlation of heart rate, corrected QT interval and hypoglycemia. However, the performance in terms of the specificity is not provided.

During hypoglycemia, the most profound physiological changes are caused by activation of the sympathetic nervous system. Among the strongest responses are sweating and increased cardiac output [2], [12], [13]. Sweating is mediated through sympathetic cholinergic fibres, while the change in cardiac output is due to an increase in heart rate and increase in stroke volume [2]. Tattersall and Gill [14] raised the possibility of hypoglycemia-induced arrhythmias; experimental hypoglycemia has been shown to prolong QT intervals and dispersion in both non-diabetic subjects and in those with type 1 and type 2 diabetes [15].

The statistical regression method [16] is a common empirical approach to develop such classification models for various medical diagnoses, such as diabetic nephropathy [17], and acute gastrointestinal bleeding [18]. However, the regression method does not perform well if the data distribution is highly irregular. Recently, computational intelligence technologies, such as fuzzy systems [19], [20], support vector machines [21], and neural networks [22], [23], have been applied to modeling and classification for medical diagnostic purposes of electrocardiogram (ECG) and electroencephalograph (EGG) classifications [24], [25], [26], [27], cardiovascular responses [28], [29], breast cancer [30], blood cells [31], skull and brain [32], dermatological disease [33], [34], gene selection [35], and heart disease [36], etc. The main feature of a fuzzy system is its decision-making ability based on the system representation provided by human experts. As a result, the output of the fuzzy system can be determined by a set of linguistic rules that can be easily understood. Neural networks are proven to be a universal approximator [37]. They have been used as classification models for medical-diagnostic purposes [38]. The advantage of using neural networks in diagnosis is their versatility in addressing both the nonlinear and human nature of the patients’ data. Support vector machines have a proven classification performance in various applications [39], such as those dealing with cardiac signals [2], [40], and are excellent in binary classification problems. Traditional optimization methods of least square algorithms and the gradient descent method have a problem of trapping in the local optima. To overcome this drawback, evolutionary algorithms, such as particle-swarm optimization (PSO) [41], genetic algorithm [42], differential evolution [43], and ant-colony optimization [44] have been introduced. These algorithms are efficient in solving the optimization problems globally. The combination of any of these computational technologies gives a good performance in biomedical applications [25], [30], [45], [46], [47], [48].

In this paper, we have developed a hybrid PSO-based fuzzy-reasoning model (FRM) for the detection of hypoglycemic episodes using physiological parameters such as heart rate and corrected QT interval of ECG signals. This FRM [19], [49], [50], [51], [52] is valuable because it reflects and incorporates expert knowledge and experience in some linguistic rules that can be easily understood by human beings. By introducing the FRM, the overall performance in terms of sensitivity and specificity of the detection system is significantly improved. To optimize the fuzzy rules and membership functions of FRM, a global learning algorithm called hybrid particle-swarm-optimization with wavelet mutation (HPSOWM) [53] is introduced. PSO is a powerful random global search technique to solve an optimization problem and it can help find the global optimal solution over a domain. In HPSOWM, wavelet mutation is introduced to overcome the drawback of possible trapping in the local optima in PSO. Furthermore, a fitness function is introduced to reduce the risk of the phenomenon of the overtraining problem [54] in this application. To realize a real clinical environment, the sampling period is shortened to 5 and 10 min and no dietary restriction is required for the patients. A study of 16 T1DM children (589 data) is given to show that the proposed method is successful in detecting nocturnal hypoglycemic episodes.

This paper is organized as follows: in Section 2, details of the development of HPSOWM-based FRM is presented. Section 3 presents the results of the detection of nocturnal hypoglycemic episodes in T1DM children. Section 4 features the conclusion.