A Clinical Decision Support System with an

J Med Syst (2014) 38:38 DOI 10.1007/s10916-014-0038-9

PATIENT FACING SYSTEMS

A Clinical Decision Support System with an Integrated EMR for Diagnosis of Peripheral Neuropathy Reeda Kunhimangalam & Sujith Ovallath & Paul K. Joseph

Received: 29 December 2013 / Accepted: 13 March 2014 # Springer Science+Business Media New York 2014

Abstract The prevalence of peripheral neuropathy in general population is ever increasing. The diagnosis and classification of peripheral neuropathies is often difficult as it involves careful clinical and electro-diagnostic examination by an expert neurologist. In developing countries a large percentage of the disease remains undiagnosed due to lack of adequate number of experts. In this study a novel clinical decision support system has been developed using a fuzzy expert system. The study was done to provide a solution to the demand of systems that can improve health care by accurate diagnosis in limited time, in the absence of specialists. It employs a graphical user interface and a fuzzy logic controller with rule viewer for identification of the type of peripheral neuropathy. An integrated medical records database is also developed for the storage and retrieval of the data. The system consists of 24 input fields, which includes the clinical values of the diagnostic test and the clinical symptoms. The output field is the disease diagnosis, whether it is Motor (Demyelinating/Axonopathy) neuropathy, sensory (Demyelinating/ Axonopathy) neuropathy, mixed type or a normal case. The results obtained were compared with the expert’s opinion and the system showed 93.27 % accuracy. The study aims at showing that Fuzzy Expert Systems may prove useful in providing diagnostic and predictive medical opinions. It enables the clinicians to arrive at a better diagnosis as it keeps the expert knowledge in an intelligent system to be used efficiently and effectively. This article is part of the Topical Collection on Patient Facing Systems R. Kunhimangalam (*) : P. K. Joseph National Institute of Technology, NIT Calicut (PO), Kozhikode, Kerala, India 673601 e-mail: [email protected] S. Ovallath Department of Neurology, Kannur Medical College, Anjarakandy (PO), Kerala, India

Keywords Fuzzy logic controller . Peripheral nerve disorders . Nerve conduction studies . Axonal neuropathy . Demyelinating neuropathy

Introduction Neurological disorders are classified according to the primary location affected viz. the peripheral nervous system (PNS) or the central nervous system (CNS). Disorders affecting the PNS consist of a spectrum of disorders out of which carpal tunnel syndrome (CTS) and symmetrical peripheral neuropathy predominates. Damage to the PNS interferes with the vital connections from the brain and the spinal cord (the CNS) to every other part of the body resulting in distortion and/or interruption in the communication between the CNS and the rest of the body. This manifests itself through symptoms ranging from temporary numbness, tingling and pricking sensations (parasthesia), sensitivity to touch, or muscle weakness to extreme symptoms like burning pain (which increases at night), muscle wasting, paralysis, or organ/ gland dysfunction [1]. More than hundred types of peripheral neuropathies have been keyed out, each with its own distinctive set of symptoms, pattern of development and prognosis. Impaired functions and symptoms depend upon the type of nerves—motor, sensory, or autonomic—that are damaged. Motor nerves control movements of muscles under our conscious control while the sensory nerves convey information from the different sense organs. Autonomic nerves regulate biological activities not under our conscious control, such as breathing, digestion, gland functions etc. [2]. Although some neuropathies may affect all three types of nerves, others primarily affect one or two types. Therefore, the patient’s condition is described by doctors using terms such as predominantly motor neuropathy, predominantly sensory neuropathy, sensory-motor neuropathy,

38, Page 2 of 14

or autonomic neuropathy. It is also important to distinguish whether the neuropathy is axonal or demyelinating in nature or both [3]. Demyelination refers to a primary pathology affecting the myelin sheaths. The loss of myelin leads to deterioration of the saltatory conduction, this result in slowing down of conduction along the nerve and may even cause conduction block. Severe demyelination may lead to secondary axonal damage. Axonal degeneration refers to a breakdown of the distal parts of axons thereby disrupting normal nerve conduction [4]. The prevalence of peripheral neuropathy in general population keeps on increasing each year with diabetes mellitus and leprosy amongst the most common etiologies [5, 6]. In developing countries such as India there is shortage of neurologists with the ratio of experts to general population being very less, so in many cases the disease goes undiagnosed. The treatment for peripheral neuropathy depends on its cause and so the initial and most crucial step in treatment is to look for the cause. Treatment should target the underlying disease process, rectify any nutritional inadequacies, and provide symptomatic treatment. Various disorders can damage peripheral nerves and cause peripheral neuropathy; so it is important to first differentiate actual neuropathy from other disorders that have a similar clinical presentation. This differentiation is best accomplished using nerve conduction studies (NCS) which can be considered as the most useful initial laboratory study in the evaluation of a patient with peripheral neuropathy [7]. NCS is a key component of electro-diagnostic evaluation, providing useful quantitative and qualitative understandings of neuromuscular functions, particularly the ability of electrical conduction of the motor and sensory nerves [8]. NCS can confirm the presence of a neuropathy [9] and provide information as to the type of fibers affected (motor, sensory, or both) and the pathophysiology (axonal loss versus demyelination). Characterization of the neuropathy helps the clinician minimize the testing needed to ascertain the etiology of the neuropathy [10]. Peripheral nerves have the ability to regenerate, provided that the nerve cell itself has not been killed. Symptoms often can be controlled, and eliminating the causes of specific forms of neuropathy often can prevent new damage while long-standing neuropathy may cause lasting nerve damage and generally reduces the chances of successful treatment [11]. All these points to the fact that the correct diagnosis at the earliest possible stage is very essential and is a crucial step in the electro-diagnosis procedure and is crucial for defining prognosis and therapeutic measures. A Clinical decision support system (CDSS) is any computer program which helps in the diagnosis of diseases. The use of mathematical sciences, engineering principles and computer technology in the diagnosis and treatment of various illnesses has highly increased nowadays [12, 13]. Despite being extremely complex and uncertain, intelligent systems such as fuzzy logic and artificial neural networks [14] have been

J Med Syst (2014) 38:38

widely employed for the development of CDSS in the field of medicine [15, 16]. These techniques have become well established as computational methodologies with potential to be effective in any field of study, especially medicine [17]. Developers of health care software have attributed improvements in patient care to these applications but as with any health care intervention, such claims require confirmation in clinical trials [18]. Fuzzy logic, a logic similar to human thinking and interpretation is highly suited and practical for developing knowledge-based systems in medicine [19] for interpretation of medical findings, diagnosis, treatment selection etc. [20]. A fuzzy expert system (FES) is a type of rule based form of artificial intelligence which uses a collection of membership functions and rules to reason about data [21]. Our aim was to develop such a FES using fuzzy logic controller which could diagnose the type of peripheral neuropathy from the clinical symptoms and the NCS data and thereby help the clinician in deciding the treatment options as soon as possible. Diagnosis or classification of a disease is often done by the specialist using a set of rules and the designed system involves the collection of these rules, together with an inference engine for evaluating the rule base for a given set of inputs. This method allows imprecision in the user inputs as well as in the rule base specification. It mimics the cognitive decision making ability of the specialist and enables the clinicians to arrive at a better diagnosis. Early detection is very crucial in neuropathy patients and it requires the assistance of a specialist for correct diagnosis, so such an expert system can be of immense use in those areas where the service of such specialists may not be readily available and it’s use is recommended to shorten the time and improve the accuracy of diagnosis in patients with suspected peripheral neuropathy.

Methods Database description In this study NCS data obtained from Kannur Medical College, Kerala, India was used to check the accuracy of the developed FES. Out of the 104 data, 20 were normal cases which had normal NCS values and had no electrophysiological evidence of any form of neuropathy, 39 cases were those of patients suffering from motor neuropathy either axonal and demyelinating type, 45 were patients suffering from sensory neuropathy either axonal or demyelinating type. The ethical committee approval was obtained. The study used the following median nerve and ulnar nerve measures: median motor and sensory latencies and amplitudes, median motor and sensory velocities, ulnar motor and sensory latencies and amplitudes, ulnar motor and sensory velocities. The attributes of the data used are given in Table 1

J Med Syst (2014) 38:38 Table 1 The attributes of the nerve conduction study dataset

Page 3 of 14, 38

Attribute no.

Attribute description

Attribute range

Mean

Standard deviation

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Motor Median latency(msec) Motor Median Amplitude Motor Median Nerve Conduction Velocity(m/s) Motor Ulnar latency(msec) Motor Ulnar Amplitude Motor Ulnar Nerve Conduction Velocity (m/s) Sensory Median latency(msec) Sensory Median Amplitude Sensory Median Nerve Conduction Velocity(m/s) Sensory Ulnar latency(msec)

2–10 4–29 30–70 2–10 3–30 30–70 2–10 5–40 30–70 2–10

4.1 9.8 54.35 3.5 8.6 53.76 3.6 15.8 49.8 3.3

3.32 5.7 10.38 1.69 4.5 8.9 1.25 5.9 9.79 1.36

11. 12.

Sensory Ulnar Amplitude Sensory Ulnar Nerve Conduction Velocity (m/s)

5–45 30–70

14.9 52.76

8.7 8.67

Nerve conduction studies assess the shape, amplitude, latency and conduction velocity of an electrical signal conducted over the tested nerve. Normal NCS results significantly decrease the likelihood of peripheral neuropathy; whereas abnormal nerve conduction findings confirm the diagnosis. The following criteria were applied for identifying the type of neuropathy [22]: If the slowing of nerve conduction affects all nerves roughly equally the diagnosis is likely to be the peripheral neuropathy. In demyelinating neuropathy the distal motor latency is prolonged and nerve conduction velocity slowed to less than 60–80 % of normal, the amplitude of the action potential usually remains unaffected. In axonal neuropathy the amplitude of the action potential is reduced, but the distal motor latency and nerve conduction velocity are somewhat unaffected. In demyelinating type of peripheral neuropathy the velocities are affected more than the amplitude and in axonal type the amplitudes are affected more than the velocity.

Procedure of NCS NCS was performed using the standard techniques with surface electrode recording on both hands of each subject using constant current stimulator. The median motor nerve at the wrist was stimulated to obtain the median motor nerve data, the recording was done over the abductor pollicis brevis muscle. The ulnar motor data was obtained by stimulating the ulnar nerve at the wrist, below the elbow, and above the elbow and recording over the abductor digiti minimi muscle. Sensory responses were obtained by applying stimulation at the wrist and recording from the index finger to get the median data and the little finger to get the ulnar data.

Design of the clinical decision support system The CDSS developed consists of three main parts: the input, the rule set and the output. In this study we have aimed at developing a MATLAB graphical user interface (GUI) system which models the reasoning process of the consultants in the particular medical scenario under consideration. The proposed system is not only used for diagnosis, but also used to store and read the results of the diagnosis for future reference. The system has three windows the first one is the SEARCH WINDOW which provides an interface where we can add the details of a new patient as well as retrieve the d at a a l r ea dy s t or ed . T h e n e x t w i n d o w is th e PERIPHERAL_NEUROPATHY_PREDICTOR window which is the interface to add the NCS values, get the diagnosis as well as add the full details including the diagnosis to the medical database. The third window is the EXISTING DETAILS window were the data stored can be retrieved and edited. The user has to enter the NCS values into the interface together with the patient details and the symptoms and when clicked on the Get Result Button the diagnosis is output by the program. The program can distinguish between and give the result as “Normal”, “Sensory (Demyelinating Type)”, “Sensory (Axonopathy)”, “Motor (Demyelinating Type)”, “Motor (Axonopathy)”, and “Mixed Peripheral Neuropathy”. The program has also provision for storing the result on a spreadsheet by clicking on the Save Result Button. This stored data can be easily retrieved for further reference as and when needed by providing the patient name and/or patient id and clicking on the Get Details Button. Design of fuzzy expert system An expert system is a computer program that helps in solving problems demanding substantial human expertness by using explicitly exhibited domain knowledge and computational

38, Page 4 of 14

decision procedures. They are designed to make available some of the skills of an expert to non experts, as they attempt to imitate the thinking patterns and logical decisions of an expert [23]. The FES makes use of the theory of fuzzy reasoning i.e. drawing conclusions from data and rules that associate data to conclusions, under conditions of uncertainty or contradictions. The process of drawing conclusions from existing data is called inference; new truths are inferred from the old ones and the conclusions reached in early stages of the reasoning process are modified or nullified as the process proceeds from one step to the next. The classical logic has only two truth values, true or false, and so the process of inference is simplified as compared to fuzzy logic, where we have to be concerned not only with propositions but also with their truth values. Every FES has a fuzzy inference system that reasons using fuzzy logic membership functions. The membership functions refer to the degree to which the value of a particular attribute belongs to a set. The FES designed and employed in this paper can be generalized by means of a simple structure as shown in Fig. 1(a). Figure 1(b) shows the block diagram of the developed system which consists of Matlab user interface and simulink blocks.

Fig. 1 Outline of the developed system and the basic fuzzy expert system a The block diagram of a basic Fuzzy Expert System b The block diagram of the developed system for the diagnosis of peripheral nerve disorders using Matlab GUI and simulink

J Med Syst (2014) 38:38

Model development The FES developed in this paper employs the Mamdani type fuzzy inference technique [24]. Fuzzy inference is the procedure of developing the mapping from a given input to an output using fuzzy logic. The mapping then delivers a foundation from which decisions can be made, or patterns can be recognized. This technique is performed in four steps viz fuzzification of the input variables, rule evaluation, aggregation of the rule outputs and defuzzification. The fuzzifier converts the crisp inputs to fuzzy inputs. Each rule is recorded by stating the membership of each input and the corresponding required output membership. The knowledge base provides input membership functions to the fuzzifier, rule-base to the inference engine, and output membership functions to the defuzzifier. Then, the fuzzy outputs are fed into the defuzzifier to generate crisp outputs. Fuzzification The fuzzifier converts the crisp inputs which are supplied to the system to fuzzy inputs and also determine the degree to

J Med Syst (2014) 38:38

which these inputs belong to each of the appropriate fuzzy sets. These fuzzy inputs are then used in the inference engine to generate fuzzy outputs. The process of knowledge acquisition is one of the most difficult and complex tasks in the construction of expert systems as eliciting knowledge from experts encounters numerous obstacles. The experts, in spite of being highly skilled in solving problems in their field, often feel difficulty in stating their knowledge in an orderly and logical manner or sometimes even in understanding their own decision making processes. For developing diagnostic tool for peripheral neuropathy, data is required that is capable of representing each type of the disease. Basically the data consists of physical signs and symptoms of patients, medical reports etc. In the present study 104 NCS reports were collected and analysed. By consulting the specialist and by analysing the data of the patient’s 20 NCS values and four symptoms were finalized as the inputs for diagnosing the severity of the disease. Then fuzzy values were assigned for each of these input variables to get different fuzzy sets based on the expertise of the specialists and knowledge from the standard textbooks. These fuzzy mapping or membership functions can have a variety of shapes such as triangular or trapezoidal shapes, depending on how the expert relates different domain values to belief values. Each of the 24 input variables were assigned membership functions each having triangular or trapezoidal shapes. The membership function plots for the input and the output variables are shown in Fig. 2 Rule determination and rule evaluation The basic requirement of rule-based systems is that the expert’s knowledge and thinking patterns should be specified in an explicit manner [25]. The set of rules in a FES is known as the rule-base or knowledge base. Obtaining this domain knowledge and writing proper rules can also be called the knowledge acquisition phase. Fuzzy rule-based systems, in addition to handling of uncertainties also have several additional capabilities. Here approximate numerical values can be specified as fuzzy numbers. Numerical input values can be easily translated into descriptive words such as “very low”, “low”, “normal”, “high”, “very high” etc. The performance of an FES mainly depends on its rule base so the optimization of the membership function distributions stored in the data base is the most important process. In an expert system the knowledge is typically laid out in the form of a set of rules. The link between the rules and the membership functions may be further stressed by the statement that membership functions have no significance without the associated rules which will be used to perform inference—i.e. the rules provide the domain setting necessary to render meaningful the linguistic variables and the membership functions [26]. The rules in a fuzzy expert system are in the form; If x is low and y is

Page 5 of 14, 38

medium, then z is high, where x and y are input variables, z is an output variable, low is a membership function (fuzzy subset) defined on x, medium is a membership function defined on y, and high is a membership function defined on z. The antecedent or the preceding part (the rule’s premise) describes the degree to which the rule applies, while the conclusion part (the rule’s consequent) assigns a membership function to each of the output variables. If a fuzzy rule has more than one antecedents, the fuzzy operators AND or OR are used to obtain a single value that represents the result of the antecedent evaluation. This truth value is then applied to the consequent membership function. Based on the descriptions of the input and output variables 105 rules were constructed by selecting an item in each input and output variable box and one connection (AND). None was chosen as one of the variable qualities to exclude any of the variables from a given rule. The weight was specified to unity. A sample rule and its interpretation is as follows Rule: If (MML(R) is NORMAL) and (MML(L) is NORMAL) and (MMA(R) is NORMAL) and (MMA(L) is NORMAL) and (MMNCV is NORMAL) and (MUL(R) is NORMAL) and (MUL(L) is NORMAL) and (MUA(R) is NORMAL) and (MUA(L) is NORMAL) and (MUNCV is NORMAL) and (SML(R) is HIGH) and (SML(L) is VERY HIGH) and (SMA(R) is NORMAL) and (SMA(L) is LOW) and (SMNCV is VERY LOW) and (SUL(R) is VERY HIGH) and (SUL(L) is HIGH) and (SUA(R) is NORMAL) and (SUA(L) is LOW) and (SUNCV is LOW) and (PAIN is SEVERE) and (NUMBNESS is SEVERE) and (PARASTHESIA is SEVERE) and (BURNING SENSATION is SEVERE) then (DIAGNOSIS is SENSORY DEMYELINATION) Interpretation: If Motor Median Latency (right) is normal and Motor Median Latency (left) is normal and Motor Median Amplitude (right) is normal and Motor Median Amplitude (left) is normal and Motor Median Nerve Conduction Velocity is normal and Motor Ulnar Latency (right) is normal and Motor Ulnar Latency (left) is normal and Motor Ulnar Amplitude (right) is normal and Motor Ulnar Amplitude (left) is normal and (Motor Ulnar Nerve Conduction Velocity is normal and Sensory Median Latency (right) is high and Sensory Median Latency (left) is very high and Sensory Median Amplitude (right) is normal and Sensory Median Amplitude (left) is low and Sensory Median Nerve Conduction Velocity is very low and Sensory Ulnar Latency (right) is very high and Sensory Ulnar Latency (left) is high and Sensory Ulnar Amplitude (right) is normal and Sensory Ulnar Amplitude (left) is low and Sensory Ulnar Nerve Conduction Velocity is low and pain is severe and numbness is severe

38, Page 6 of 14

J Med Syst (2014) 38:38

Fig. 2 Membership function plots for the various input and output variables a Sensory Ulnar NCV b Sensory median NCV c Motor Ulnar NCV d Motor median NCV e Sensory Ulnar Latency f Sensory Ulnar Amplitude g Sensory median latency h Sensory median Amplitude i

Motor Ulnar latency j Motor Ulnar Amplitude k Motor median latency l Motor median Amplitude m burning sensation n Parasthesia o Pain p Numbness q Output Variable-Disease diagnosis

and parasthesia is severe and burning sensation is severe then diagnosis is sensory demyelination.

output variable. The inference methodology used is the Mamdani inference method [27]. In Mamdani inference method rules are of the following form:

Aggregation of the rule outputs It is the process of the unification of the rules. The membership functions of all the rule consequents previously clipped during rule evaluation are taken and combined into a single fuzzy set. In this process a number of clipped consequent membership functions are changed into one fuzzy set for each

Ri : if x1 is Ai1 and . . . and xr is Air then y is Ci for i=1, 2, . . ., L; where L is the number of rules, xj (j=1, 2, . . ., r) are the input variables, y is the output variable, and Aij and Ci are fuzzy sets that are characterised by membership functions Aij (xj) and Ci (y), respectively. The consequence of each rule is characterised by a fuzzy set Ci. The final output of a Mamdani system is one or more arbitrarily

J Med Syst (2014) 38:38

Page 7 of 14, 38

Fig. 2 (continued)

complex fuzzy sets which usually need to be defuzzified. Mamdani-type inference, as defined for the toolbox, expects the output membership functions to be fuzzy sets. After the aggregation process, there is a fuzzy set for each output variable that needs defuzzification.

Defuzzification of the output Though the concept of fuzziness helps the rule evaluation during the intermediate steps, the final desired output for each variable is generally a single number i.e. a crisp value. However, the aggregate of a fuzzy set constitutes a range of output

38, Page 8 of 14

values, and so it must be defuzzified in order to resolve a single output value from the set [28]. The defuzzification method used here was the Centroid calculation, which returns the center of area under the curve. Every rule was examined for a given set of input values using the AND operation and the rule which satisfied the operational logic was used to generate the output for the inference system. The output given by each rule was aggregated and then defuzzified using centroid calculation to generate a single output which was a single number representing the diagnosis. After setting up the fuzzy inference system (FIS) the next step was building the system with Fuzzy Logic Controller (FLC) with Rule Viewer block. This implements the FIS with the Rule Viewer in simulink. Once we create the fuzzy system we can readily embed our system directly into a simulation and integrate it with the FIS. For the Mamdani FIS, the FLC block automatically generates a hierarchical block diagram representation of the FIS. This automatic model generation ability is called the Fuzzy Wizard. The FLC with Rule Viewer block is an extension of the Fuzzy Logic Controller block. It allows us to visualize how rules are fired during simulation.

Results The performance of the CDSS in diagnosis of peripheral nervous disorders The GUI developed for the PERIPHERAL NEUROPATHY PREDICTOR is shown in Fig. 3. The program front panel consists of an interface for SEARCH as shown in Fig. 3(a) here giving the name of the patient and clicking on the Search Button gives the details of the patient, if the name does not match it gives the names of all the most probable matches. To get the full details of the patient click on the patient name and click the Get Results Button, this opens up the interface EXISTING_DETAILS as shown in Fig. 3(c). It gives the entire stored data of the particular patient; there is also provision to edit the details of the patient from this interface. To add a new patient’s data and also for using the CDSS for disease prediction clicking on the New Patient Button on the SEARCH panel opens up the PERIPHERAL_NEUROPATHY_PREDICTOR window. The patient details and the NCS parameters are entered into it Fig. 3(b) show the interface with the details of a patient suffering from sensory demyelinating type neuropathy. The user inputs the various NCS values together with the degree of the symptoms into the interface and the result could b e g o t . O n c l i c k o f t h e G e t R e s u l t B u tt o n t h e PERIPHERAL_NEUROPATHY_PREDICTOR predicts the type of neuropathy and on click of the Save Result Button it stores all data in the database. This data can be retrieved for future use giving the patients name and/or ID. The program gave accurate results which were in agreement with the

J Med Syst (2014) 38:38

specialist’s diagnosis. Whenever the data entered was not within the designated limits for the disorders the program gave the result as Check the Data Entered. So the system is able to identify the uncertainty cases and this helps the physician to give a special consideration to deal with them which results in a better management of the disease. Figure 4 shows the underlying Fuzzy expert system which consists of the fuzzy logic controller with rule viewer. There is a subsystem Diagnosis block along with the FLC which consists of an embedded Matlab program and other blocks along with the display blocks. The GUI and the FLC is linked when the Get Result Button on the PERIPHERAL_NEUROPATHY_PREDICTOR interface is clicked. The NCS clinical values and the symptom degree values input into the GUI gets automatically transferred to the FES and get entered into the FLC and the FES is run automatically. The Diagnosis block consists of an embedded Matlab function block which does a maximization function and returns the value which has the highest membership function. The result is got and is displayed in the diagnosis block as shown in Fig. 4(a), when the simulation is run the diagnosis appears as display along with its membership function. For the given inputs in Fig. 4(a) the result is shown as ‘Sensory(Demyelinating Type)’ in the Diagnosis block and the membership function shown is 0.8533 which means the diagnosis is sensory demyelinating type neuropathy with a membership function 0.8533. Based on the rules the inference system made the diagnosis by following AND connection and then defuzzyfying the generated output using the centroid method. Figure 4(c) gives the defuzzification module, it uses the center of area method for defuzzification and Fig. 4(d) gives one of the rule blocks. In fuzzy logic the truth of a statement is matter of degree so the AND connection performed a min operation. Based on the AND operation every rule was examined for a given set of input values and the rule which satisfied the operational logic was used to generate the output for the inference system. Based on these rules the roadmap of the whole FIS rule viewer is obtained as shown in Fig. 5 Rule Viewer shows the active rules, the individual membership functions and how they are influencing the results. It displays a guideline of the entire fuzzy inference process. In Fig. 5 the 25 plots shown represents the antecedents and consequent of each rule. Each rule consists of a row of plots, and each column gives the value of a particular variable. On the left of each row we can see the rule numbers displayed. The first 24 columns of plots show the if-part of each rule and the last column of plots show the then-part of each rule. The plots which are blank in the if-part of any rule represent the depiction of ‘none’ for the variable in the rule. The last plot in the twenty fifth column corresponds to the aggregate weighted decision for the given inference system and it depends upon the input values of the system. Though during the intermediate

J Med Syst (2014) 38:38

Page 9 of 14, 38

Fig. 3 The GUI of the developed CDSS a The SEARCH interface window b The PERIPHERAL_ NEUROPATHY_PREDICTOR interface window c The EXISTING_DETAILS interface window

steps of rule evaluation we are dealing with fuzzy values, the final output for each variable is in a crisp form. So the

aggregate of the fuzzy sets are defuzzified in order to resolve upon a single output value from the set. The defuzzification

38, Page 10 of 14

J Med Syst (2014) 38:38

Fig. 4 The fuzzy expert system developed in simulink a the FES with the FLC with rule viewer block, the subsystem block labeled Diagnosis displays the result of the diagnosis, the display block labeled membership

function gives the membership function of the diagnosis b the subsystem Diagnosis with an embedded MATLAB function c the defuzzification block which uses the center of area method d a rule block

method used is the centroid method, and it returns the center of area under the curve which is displayed as a bold vertical line on this plot. On the topmost part above each column, the current values of each of the input variables are displayed. The

variables and their current values are displayed on top of the columns. The Rule Viewer provides us with a visual display of the interpretation of the entire fuzzy inference process and it also expresses how the shape of certain membership functions

Fig. 5 The rule viewer

J Med Syst (2014) 38:38

Page 11 of 14, 38

determines the overall result. The Rule Viewer can be looked upon as a sort of micro view of the FIS since it shows one computation at a time, in great detail. Using the developed FES 104 NCS data were tested which had normal cases, sensory demyelinating, sensory axonopathy, motor demyelinating and motor axonopathy cases. The accuracy of the system was thus found out with the consultation of the expert, the FES showed a fairly good accuracy of 93.26 % as shown in Table 2(a). The test performance of the classifiers was determined by the computation of specificity, sensitivity, positive predictive value and negative predictive value which are defined as follows. Specificity is the ratio of the number of true negative decisions to the number of actual negative cases; sensitivity is the ratio of the number of true positive decisions to the number of actual positive cases; positive predictive value gives how likely the patient has the disease given the test is positive and the negative predictive value gives how likely the patient does not have the disease given the test is negative [29]. The values for these statistical variables are given in Table 2(b). It can be seen that the system gave good results as far as the statistical performance were considered.

Discussion Diagnosis and management of diseases is indeed a difficult task which can be mastered slowly through years of observation and experience as well as thorough subject knowledge. This is due to the fact that most of the clinical scenarios have a vagueness varying in degree associated with them. Medical problems, cannot be generalized or analyzed using binary logic i.e. with a ‘yes’ or a ‘no’ and an analytical program is required. Fuzzy logic, which has the ability of merging human heuristics into computer-assisted decision making, provides a good solution to the problem. This work was undertaken with an aim to design an expert system for the diagnosis of peripheral nerve disorders using Fuzzy Logic which will be helpful for the patient to take proper curative measures before the severity increases. The results obtained from the system reveals that the diagnostic system is giving expected results and its efficacy has been endorsed by specialist doctors in the field. The system developed was not meant to replace the specialist, yet it can be used to assist a general practitioner or specialist in diagnosing and predicting patient’s condition. Employing the use of computer-aided techniques in medical applications could

Table 2 The statistical parameters of the developed system

a

Actual Diagnosis

Predicted Diagnosis

Sensory (Demyelinating) Sensory (Axonopathy) Motor (Demyelinating) Motor (Axonopathy) Normal

Sensory (Demyelin ating)

Sensory (Axonopathy)

Motor (Demye linating )

Motor (Axonopathy)

Normal

19

1

0

0

0

20

1

23

0

0

1

25

0

0

15

0

1

16

1

0

0

21

1

23

0

0

1

19

20

24

15

22

22

93.26%

0 21

b

Normal Sensory (Demyelinating) Sensory (Axonopathy) Motor (Demyelinating) Motor (Axonopathy)

Positive Predictive Value 95%

Negative Predictive Value 96.4%

98.7%

95%

96.4%

95.8%

96.25%

93.75%

96.25%

99%

97.7%

91.3%

96.6%

86.36%

98.7%

95%

96.4%

Sensitivity 86.36%

Specificity 98.7%

90.4%

38, Page 12 of 14

reduce the cost, time, human expertise and medical error [30]. Such systems have been shown to improve prescribing practices, reduce serious medication errors and enhance the delivery of preventive care services [31]. In the arena of medical diagnosis it acts as a powerful tool to help doctors to examine and model clinical data and make use of them for a number of medical applications. The lack of adequate number of specialists often is a grave problem which leads to misdiagnosis and in many a case the disease goes undiagnosed. The significance of the developed CDSS lies on the fact that such system can effectively assist the clinicians to arrive at a more accurate diagnosis even in the absence of an experienced specialist/neurologist. CDSS has been successfully used in many areas of medicine such as breast cancer diagnosis and classification [32], for diagnosis of coronary heart disease [33] using a fuzzy rulebased system with multi objective genetic algorithm. The work in [34] mathematically models the variability of human decision making process using type-2 fuzzy sets, it attempts to improve the accuracy of a fuzzy expert decision making system by tuning the parameters of type-2 sigmoid membership functions of fuzzy input variables and hence determining the most appropriate type-1 membership function. In [35] a system was designed for the diagnosis of Arthritis using fuzzy logic controller and in [36] a new hybrid case-based reasoning approach for medical diagnosis systems is proposed which integrates case-based reasoning and rule-based reasoning, and also applies the adaptation process automatically by exploiting adaptation rules. In [37] a methodology for automatic detection of epileptic conditions from EEG signals were suggested, four entropy features were extracted and fed to different classifiers and the result showed that the fuzzy classifier was able to differentiate the cases with the highest accuracy of 98.1 %. In the paper subtractive clustering method was used for the generation of the FIS but in our paper the FIS rules were laid down after consulting the experts and the standard text books. For assessment of severity of asthma a rule based expert system was presented in [21], the knowledge base of this system has been designed as modular were every module can be used independently for evaluation of indicators. This system can be used without requiring of lab data (lung function test, and atopy test), so it can be applied in primary care setting for assessment severity of asthma by general physician. However the result of system without laboratory data may not have the precision when all variables have been considered for evaluating severity. In our system we have taken into consideration both the lab data as well as the symptoms. A system for cardiac health diagnosis using fuzzy logic based data fusion was presented in [38], an accuracy of 93 % was obtained for the developed system. In the paper fuzzy logic decision function was created by fuzzyfying Boolean rules and assigned with a probability for the patient’s state. In our paper the fuzzy rules were directly written from medical knowledge in different forms. A CDSS for prevention of venous thromboembolism

J Med Syst (2014) 38:38

was introduced in [39] while in [40] an adaptive neuro-fuzzy network is used to classify heart abnormalities. Detection of seizure activity using a two-layered Learning Vector Quantization neural network was done in [14]. There have been a number of studies to systematically review controlled clinical trials assessing the effects of computer-based CDSS on physician performance and patient outcomes [41]. In comparison to applicability of fuzzy logic in medicine the concept is still new in the field of neurosciences [42]. This was clearly highlighted from the fact that the contribution to the literature on fuzzy logic was much less from neurosciences compared to other disciplines of medicine. In our recent works we had shown how artificial neural networks and fuzzy logic could be used successfully for the diagnosis of peripheral nerve disorders such as CTS [43] and neuropathy [44, 45]. One drawback of such studies, which makes use of the NCS data are the inherent shortcomings of the interpretation of the results, which include lack of standardization and absence of population-based reference intervals. A potential limitation of electro diagnostic studies is that they are able to test only the large, myelinated nerve fibers. This limits their sensitivity in detecting neuropathies of the small nerve fibers (i.e., those with pain, temperature, and autonomic functions). There is no reliable means of studying proximal nerves. NCS results can be normal in patients with small-fiber neuropathies, and lower extremity sensory responses can be absent in normal elderly patients.

Conclusion In this paper, we propose a novel and efficient clinical decision support system using fuzzy logic for the identification of peripheral nerve disorders. The developed system also consists of an integrated electronic health record system which can be used to store the results and also to retrieve them for future reference. The system was tested by consulting the expert and the system showed good accuracy. The developed system can be made more foolproof by testing the system on a larger cohort and improvising upon the rule base if needed. The same system can be developed to identify other types of peripheral nerve disorders including CTS by additions to the knowledge base and rule base. However, given the limited range of clinical settings in which they have been tested, such systems must be evaluated rigorously before widespread introduction into clinical practice. Thus we conclude that studies involving the use of such a Fuzzy Expert System in providing diagnostic and predictive medical opinions are highly promising for the future. They can add value if embedded into the routine clinical consultations and used judiciously but can never completely replace the clinician.

Conflicts of interest The authors declare that they have no conflicts of interest.

J Med Syst (2014) 38:38

References 1. Woolf, C. J., and Mannion, R. J., Neuropathic pain: Aetiology, symptoms, mechanisms, and management. Lancet 353(9168): 1959–1964, 1999. 2. Boulton, A. J. M., Gries, F. A., and Jervell, J. A., Guidelines for the diagnosis and outpatient management of diabetic peripheral neuropathy. Diab. Med. 15(6):508–514, 1998. 3. Martyn, C. N., and Hughes, R. A., Epidemiology of peripheral neuropathy. J. Neurol. Neurosurg. Psychiatry 62(4):310–318, 1997. 4. Spencer, P.S., Miller, M.S., Ross, S.M., Schwab, B.W., and Sabri, M.I., Biochemical mechanisms underlying primary degeneration of axons. Alterations of Metabolites in the Nervous System. Springer. 31–65, 1985. 5. Jensen, P. G., and Larson, J. R., Management of painful diabetic neuropathy. Drugs Aging 18(10):737–749, 2001. 6. Van Brakel, W. H., Peripheral neuropathy in leprosy and its consequences. Lepr. Rev. 71(Suppl):S146–S153, 2000. 7. Behse, F., Buchthal, F., and Carlsen, F., Nerve biopsy and conduction studies in diabetic neuropathy. J. Neurol. Neurosurg. Psychiatry 40(11):1072–1082, 1977. 8. Chemali, K. R., and Tsao, B., Electrodiagnostic testing of nerves and muscles: When, why, and how to order. Cleve. Clin. J. Med. 72(1): 37–48, 2005. 9. Perkins, B. A., Olaleye, D., Zinman, B., and Bril, V., Simple screening tests for peripheral neuropathy in the diabetes clinic. Diabetes Care 24(2):250–256, 2001. 10. Cheah, B. C., and Kiernan, M. C., Neurophysiological methodologies: Diagnosis of peripheral nerve disease and assessment of pharmacological agents. Curr. Opin. Investig. Drugs 11(1):72– 79, 2010. 11. Tesfaye, S., Boulton, A. J., Dyck, P. J., et al., Diabetic neuropathies: Update on definitions, diagnostic criteria, estimation of severity, and treatments. Diabetes Care 33(10):2285–2293, 2010. 12. Musen, M.A., Shahar, Y., and Shortliffe, E.H., Clinical decisionsupport systems. Biomedical informatics. Springer New York: 698– 736, 2006. 13. Faust, O., Acharya, U. R., and Tamura, T., Formal design methods for reliable computer-aided diagnosis: A review. IEEE Rev. Biomed. Eng. 5:15–28, 2012. 14. Pradhan, N., Sadasivan, P. K., and Arunodaya, G. R., Detection of seizure activity in EEG by an artificial neural network: A preliminary study. Comput. Biomed. Res. 29(4):303–313, 1996. 15. Forsstrom, J. J., and Dalton, K. J., Artificial neural networks for decision support in clinical medicine. Ann. Med. 27(5):509–517, 1995. 16. Phuong, N. H., and Kreinovich, V., Fuzzy logic and its applications in medicine. Int. J. Med. Informat. 62(2):165–173, 2001. 17. Miller, R. A., Medical diagnostic decision support systems-past, present, and future a threaded bibliography and brief commentary. J. Am. Med. Inform. Assoc. 1(1):8–27, 1994. 18. Garg, A. X., Adhikari, N. K., McDonald, H., Rosas-Arellano, M. P., Devereaux, P. J., Beyene, J., Sam, J., and Haynes, R. B., Effects of computerized clinical decision support systems on practitioner performance and patient outcomes: A systematic review. J. Am. Med. Assoc. 293(10):1223–1238, 2005. 19. Zadeh, L. A., The role of fuzzy logic in the management of uncertainty in expert systems. Fuzzy Sets Syst. 11(1):197–198, 1983. 20. Ramnarayan, P., and Britto, J., Paediatric clinical decision support systems. Arch. Dis. Child. 87(5):361–362, 2002. 21. Zolnoori, M., Zarandi, M. H., Moin, M., and Teimorian, S., Fuzzy rule-based expert system for assessment severity of asthma. J. Med. Syst. 36(3):1707–1717, 2012. 22. Kimura, J., Principles and pitfalls of nerve conduction studies. Ann. Neurol. 16(4):415–429, 1984. 23. Buckley, J. J., Siler, W., and Tucker, D., A fuzzy expert system. Fuzzy Set Syst. 20(1):1–16, 1986.

Page 13 of 14, 38 24. Keshwani, D. R., Jones, D. D., Meyer, G. E., and Brand, R. M., Rulebased Mamdani-type fuzzy modeling of skin permeability. Appl. Soft Comput. 8(1):285–294, 2008. 25. Clancey, W. J., The epistemology of a rule-based expert system-a framework for explanation. Artif. Intell. 20(3):215–251, 1983. 26. Wang, L. X., and Mendel, J. M., Generating fuzzy rules by learning from examples. IEEE Trans. Syst. Man Cybern. Syst. Hum. 22(6): 1414–1427, 1992. 27. Ting, Y., Lu, W. B., Chen, C. H., and Wang, G. K., A fuzzy reasoning design for fault detection and diagnosis of a computer-controlled system. Eng. Appl. Artif. Intell. 21(2):157–170, 2008. 28. Saade, J. J., and Diab, H. B., Defuzzification techniques for fuzzy controllers. IEEE Trans. Syst Man. Cybern.B Cybern 30(1):223–229, 2000. 29. Brenner, H. E. R. M., and Gefeller, O. L. A. F., Variation of sensitivity, specificity, likelihood ratios and predictive values with disease prevalence. Stat. Med. 16(9):981–991, 1997. 30. Bright, T. J., Wong, A., Dhurjati, R., Bristow, E., Bastian, L., Coeytaux, R. R., Samsa, G., Hasselblad, V., Williams, J. W., Musty, M. D., Wing, L., Kendrick, A. S., Sanders, G. D., and Lobach, D., Effect of clinical decision-support systems: A systematic review. Ann. Intern. Med. 157(1):29–43, 2012. 31. Kawamoto, K., Houlihan, C. A., Balas, E. A., and Lobach, D. F., Improving clinical practice using clinical decision support systems: A systematic review of trials to identify features critical to success. Br. Med. J. 330(7494):765, 2005. 32. Ganesan, K., Acharya, R. U., Chua, C. K., Min, L. C., Mathew, B., and Thomas, A. K., Decision support system for breast cancer detection using mammograms. Proc. Inst. Mech. Eng. H. 227(7): 721–732, 2013. 33. Lahsasna, A., Ainon, R. N., Zainuddin, R., and Bulgiba, A., Design of a fuzzy-based decision support system for coronary heart disease diagnosis. J. Med. Syst. 36(5):3293–3306, 2012. 34. Das, S., Chowdhury, S. R., and Saha, H., Accuracy enhancement in a fuzzy expert decision making system through appropriate determination of membership functions and its application in a medical diagnostic decision making system. J. Med. Syst. 36(3):1607–1620, 2012. 35. Singh, S., Kumar, A., Panneerselvam, K., and Vennila, J. J., Diagnosis of arthritis through fuzzy inference system. J. Med. Syst. 36(3):1459–1468, 2012. 36. Sharaf-El-Deen, D. A., Moawad, I. F., and Khalifa, M. E., A new hybrid case-based reasoning approach for medical diagnosis systems. J. Med. Syst. 38(2):1–11, 2014. 37. Acharya, U. R., Molinari, F., Sree, S. V., Chattopadhyay, S., Ng, K. H., and Suri, J. S., Automated diagnosis of epileptic EEG using entropies. Biomed. Signal Proc. Contr. 7(4):401–408, 2012. 38. Kannathal, N., Acharya, U. R., Ng, E. Y. K., Krishnan, S. M., Min, L. C., and Laxminarayan, S., Cardiac health diagnosis using data fusion of cardiovascular and haemodynamic signals. Comput. Methods Appl. Mech. Eng. 82(2):87–96, 2006. 39. Durieux, P., Nizard, R., Ravaud, P., Mounier, N., and Lepage, E., A clinical decision support system for prevention of venous thromboembolism. J. Am. Med. Assoc. 283(21):2816–2821, 2000. 40. Kannathal, N., Lim, C. M., Rajendra Acharya, U., and Sadasivan, P. K., Cardiac state diagnosis using adaptive neuro-fuzzy technique. Med. Eng. Phys. 28(8):809–815, 2006. 41. Hunt, D. L., Haynes, R. B., Hanna, S. E., and Smith, K., Effects of computer-based clinical decision support systems on physician performance and patient outcomes: A systematic review. J. Am. Med. Assoc. 280(15):1339–1346, 1998. 42. Godil, S. S., Shamim, M. S., Enam, S. A., and Qidwai, U., Fuzzy logic: A “simple” solution for complexities in neurosciences? Surg. Neurol. Int. 2:24, 2011. 43. Kunhimangalam, R., Ovallath, S., and Joseph, P. K., A novel fuzzy expert system for the identification of severity of carpal tunnel syndrome. Biomed. Res. Int. 2013:846780, 2013.

38, Page 14 of 14 44. Kunhimangalam, R., Ovallath, S., and Joseph, P. K., Computer aided diagnostic problem solving: Identification of peripheral nerve disorders. IRBM. 34(3):244–251, 2013.

J Med Syst (2014) 38:38 45. Kunhimangalam, R., Ovallath, S., and Joseph, P.K., Artificial Neural Networks in the Identification of Peripheral Nerve Disorders. J. Mech. Med. Biol. 12(04):1240018(1–14), 2012.