Using hierarchical soft computing method to discriminate microcyte anemia
Introduction
The microcytic anemia is the most common hematological disorder encountered in Taiwan, over one billion people are anemic all over the world. Anemia, the red blood cell mass lead to the hemoglobin diminishes the capability of oxygen carrying, which would cause to death if anemic disorder were developed to acute anemia rapidly (D'Aquila et al., 2003, Rosner and Gunwald, 1997). The symptoms in patients with anemia are those of underlying disease present fatigue, syncope, dyspnea, angina pectoris, and transient cerebral, etc. (Rosner & Gunwald, 1997).
In approaching the evaluation of anemia, physicians depend on CBC to find anemia that hemoglobin (HGB) and hematocrit (HCT) levels below the normal reference range. It is hard to distinguish IDA and THA because of the feature that IDA is similar to THA. Both of IDA and THA were characterized to microcyte anemia with mean cell volume (MCV) is less than 80 fL, for discriminating effectively, the patient must do others examination the further. The classification of microcytic anemia for CBC-based differential diagnosis is considered as uncertain an approximate; for example, a patient's CBC with MCV equal to 70, HGB equal to 11, and HCT equal to 36 could be misclassified as IDA.
The reading of RBC level depends on different parameters that were provided from the automatic analyzer. Some analyzers provide the extra-parameters such as RDW standard deviation (SD) and coefficient of variation (CV), these extra-parameters were adopted by priori literature, and verified in sensitivity and specificity for microcyte anemia diagnosis (Akai et al., 1998). Since 1973, Mentzer (1973) proposed discriminant function (DF) MCV/RBC less than 13 to identify THA. The uncertainty and approximate of DF which is explained in sensitivity and specificity, subsequently, more discriminant functions was proposed. So far, most of the DFs were formulated with RBC and other items of CBC to calculate outcome, which used for microcytic anemia classification. England et al. proposed formulation MCV–RBC-(5xHGB)-k in 1979 (England & Fraser, 1979), k denotes constant adjusted by different population. Shine and Lal (1977) proposed MCV squared multiply mean red cell hematocrit (MCH). These DFs can identify only either IDA or THA has either higher sensitivity or higher specificity, due to the result of these DFs that have many false positive or false negative (Drews, 2003).
Some researchers applied the traditional statistic model such as maximum likelihood via expectation-maximization and Bayes rule (Cadez et al., 2001, McLaren et al., 2001) in which belong to hard computing. However, recently, the artificial intelligent soft computing method obtained more attentions to solved relative problem of classification (Bezdek, 1973, Birndorf et al., 1996, Markey et al., 2003, Valafar et al., 2000). Artificial neural network (ANN), fuzzy and neural–fuzzy belong to soft computing which had been employed to aid with physicians for clinical diagnosis (Bela et al., 2002, DeLeo and Rosenfeld, 2001, Markey et al., 2003, Woods and Bowyer, 1997, Virant-Klun and Virant, 1999). Soft computing method, aims at medical diagnosis under consideration of clinicians, is more fitting than traditional hard computing.
In this study, we proposed an architecture that combines some soft computing methods to classified microcyte anemia. In general, the clinical anemia diagnosis procedures cannot discriminate disease more detailed when the pathology reveal the MCV value less than 80 fL (Lee et al., 2003, Rosner and Gunwald, 1997). The single classifier of soft computing was employed in medicine (Bela et al., 2002, Markey et al., 2003), however not all of debatable problem of diagnosis which could be classified in single classifier. It is necessary to develop a heuristic and adaptive architecture for complex problem such as microcytic anemia diagnosis. To overcome the drawback of ambiguous boundary of diagnosis, thus we introduce soft computing hierarchical method based on the advantage of artificial intelligent, named hierarchical soft computing. Every procedure adopts one of soft computing methods, which depends on both expert knowledge and characteristics of CBC to finds the rule. HSC architecture is an adequate solution for microcyte anemia diagnosis, which is better than traditional method, and extends the scope of the anemia diagnosis.
This paper organized five sections: Section 2 introduce the material and propose a hierarchical soft computing derived from soft computing methods; Section 3 verify the result of procedures of HSC; in Section 4, we compare the performance between HSC and DFs by area under ROC curve (AUC), and accurate rate in different methods, finally, make a conclusion and discussion to end this paper.
Section snippets
Material and methods
The CBC items is listed in Table 1, and the abbreviations of them often seen in medicine or referred to physicians usage following: HGB, HCT, MCH, MCHC, MCV, PLT, RBC, WBC and RDW. Some of CBC items had interesting for discriminating anemia (Akai et al., 1998, Cadez et al., 2001, Rosner and Gunwald, 1997). In this study, the database is consisted of the CBC of anemia patient who had been documented and confirmed international statistical classification of diseases (ICD) code. The technicians
Data analysis and results
The result of HSC highlights the accuracy of every procedure that will impact the outcome of final inference system. First, we pay attention to achievements of the two classifiers, respectively, and demonstrate optimal performance of the two classifiers. Secondly, the ANFIS after trained completely, if the error validation is not small enough to satisfy further investigation. In order to eliminate this disadvantage in HSC, as possible as we can to compare the different parameters, such as
Comparison
Since a serial results verified following the HSC procedure, we attempt to compare performance of some methods, and then chose an optimal method to extract feature. We demonstrate two comparisons in this section. One is to compare these methods with specific criteria to found which has higher accuracy another is to compare with DFs of prior studies proposed.
Conclusions and discussion
There were a series of literatures applied expectation-maximum likelihood estimation (EM) to investigate the problem of classification. The EM used to classify multivariate data such as CBC, which had been obtained achievements. However, these methods have not been developed a rule-based scheme for acquiring capabilities of learning, or extract rule through the progress of training yet. In this paper, we have been demonstrated the proposed architecture, which improve the performances of single
References (28)
- et al.
An expert system to diagnose anemia and report results directly on hematology forms
Computers and Biomedical Research
(1996) - et al.
Discrimination between iron-deficiency and heterozygous-thalassemia syndromes in differential diagnosis of microcytosis
Lancet
(1979) - et al.
Self-organizing map for cluster analysis of a breast cancer database
Artificial Intelligence in Medicine
(2003) Differentiation of iron deficiency from thalassemia trait
Lancet
(1973)- et al.
The patient with anemia
Hematology Update for Psychiatrists
(1997) - et al.
A strategy to detect beta-thalassemia minor
Lancet
(1977) - et al.
Iron deficiency and-or thalassemia trait
Lancet
(1973) - et al.
Predicting the effectiveness of hydroxyurea in individual sickle cell anemia patients
Artificial Intelligence in Medicine
(2000) - et al.
Fuzzy logic alternative for analysis in the biomedical sciences
Computers and Biomedical Research
(1999) Fuzzy sets
Information and Control
(1965)
A novel kernelized fuzzy C-means algorithm with application in medical image segmentation
Artificial Intelligence in Medicine
A study of β thalassemia screening using an automated hematology analyzer
Sysmex Journal Interational
Automatic detection of distorted plethysmogram pulses in neonates and paediatric patients using an adaptive-network-based fuzzy inference system
Artificial Intelligent in Medicine
Cited by (12)
A comprehensive evaluation of the most suitable HVAC system for an industrial building by using a hybrid building energy simulation and multi criteria decision making framework
2021, Journal of Building EngineeringCitation Excerpt :Relative humidity (CE2) criterion reflects the system's ability to maintain humidity at preferred level. In addition to its effects on the human health [79] this issue is especially important in industrial buildings where the quality of the product is affected by the high humidity [78]. Visual impact (CE3) is related with dry-eye problem because of the directly contacting air flow generated by the system [81].
Classifying anemia types using artificial learning methods
2021, Engineering Science and Technology, an International JournalCitation Excerpt :In a study conducted in 2005, Yeh and Cheng proposed a solution to this problem by using the hierarchical software calculation technique of a rule-based software method. They achieved 96% accuracy on 50 samples and reported that their approach was more successful than traditional methods [27]. Allahverdi et al. published a study using the Takagi-Sugeno type neural-fuzzy (neuro-fuzzy) network method to determine childhood anemia.
A Takagi-Sugeno type neuro-fuzzy network for determining child anemia
2011, Expert Systems with ApplicationsCitation Excerpt :FCM algorithm has been used and 97% of classification accuracy has been obtained. Yeh and Cheng (2005) proposed hierarchical soft computing fitting for discriminating microcyte anemia. This study achieved sensitivity and the specificity is 90.1% and 95.8%, respectively.
A Drowsiness Detection Decision Support System using Self-Organising Map
2022, Proceedings of the 2022 IEEE Nigeria 4th International Conference on Disruptive Technologies for Sustainable Development, NIGERCON 2022Recent studies and research on sickle cell disease: Statistical analysis and machine learning approach
2020, Handbook of Research on Advancements of Artificial Intelligence in Healthcare Engineering