Professional Documents
Culture Documents
1 Introduction
Disease diagnosis is an important process in healthcare. This essential process is
conducted by physicians before giving medical treatments to the patients. The
diagnosis is often based on results of blood tests, urine tests and so forth. In cases that
required an examination of internal organs of the patient, physician would refer them
to radiologists to acquiring vital images using devices such as MRI, CT scan, x-ray, or
mammograms, depending on the organ and the suspected disease. In the past, the
radiologists or the physicians undertook themselves the reading and interpreting these
images. Later, computer aided diagnosis (CAD) was introduced in aid of diagnosing
the complex cases. However, the images produced by imaging devices such as for
breast cancer, could be affected by noise and distortion which consequently affect the
A. Abd Manaf et al. (Eds.): ICIEIS 2011, Part II, CCIS 252, pp. 128139, 2011.
Springer-Verlag Berlin Heidelberg 2011
Hybrid Intelligent System for Disease Diagnosis 129
diagnosis of the radiologists [1]. The noises or distortions in the image could also
cause errors during the diagnosis procedure. For instance, by missing tumors or
wrongly detecting tumors that in fact are non presence (false positives). These
mistakes would certainly influence the final assessment and lead to the wrong
treatment / therapy.
CAD systems such as [2-4] are utilized to support the decision of radiologist in
diagnosing images. Such systems have been reported to increase the performance of
detecting cancer by 19.6% compared to the sole diagnosis taken by a radiologist [5].
CAD systems highlight certain regions that enclose cancer or any sort of
abnormalities that might be overlooked by radiologists. Computer aided diagnosis
systems are often built with algorithms based on artificial neural networks, support
vector machines classifiers and others [3]. However, apart from the success of CAD
systems application, the issue of finding subtle cancerous cells remains cumbersome
as these cells are very difficult to be identified. In addition, [6] reported that some
radiologists tend to pay no attention to detected cancers highlighted by CAD systems.
Those radiologists might have had experiences with CAD systems that wrongly
diagnosed cases which highly affected their trust towards these systems. In this
regard, systems capable to produce accurate detection results (especially for hidden
cancerous cells) are needed.
This paper presents an architecture of a CAD system that utilizes soft
computing technique to analyze images and identify abnormalities. This
architecture combines artificial neural networks, fuzzy logic and genetic
algorithms for the learning and classification tasks. Section 2 presents some related
work in this area. Section 3 explains the proposed architecture and the hybrid
learning algorithm, and section 4 discusses an experiment to evaluate the hybrid
learning algorithm. Finally, the conclusion and a discussion on future work
directions are presented in section 5.
2 Related Work
system was based on 2D images and as some abnormalities were hidden, a 3D image
was needed to enable the detection. A system introduced by Raja et al. [11] is aimed
for a computer aided system for early diagnosis of Alzheimer disease based on single-
photon emission computed tomography (SPECT) image feature selection and random
forest classifier. The project achieved 96% classification accuracy. The above
applications are considered successful in the area of computer aided systems CAD to
diagnose different diseases. However, diagnosing subtle cancer cells is still an issue
that need to be addressed by developing better algorithms capable of revealing these
cells.
Soft computing aims at solving uncertain, imprecise or partially true problems [12].
Fuzzy Logic (FL), Neural Network (NN) and Genetic Algorithm (GA) etc. are some
techniques for soft computing. The idea of hybrid intelligent systems is to combine
more than one methodology to complement each other [13]. Nazmy et al. [14] for
instance developed an adaptive neuro-fuzzy inference system (ANFIS) for
classification of ECG signals. The performance of this classifier has been tested and
the result achieved a rate of 97% accuracy. Raja et al. [15] meanwhile developed a
computer aided diagnosis system using NN-FL hybrid technique for diagnosing
ultrasound kidney images. Another application of NN-FL technique in detecting
tumors in medical images was developed by [16]. This system used a hybrid
combination of NN-FL with expert system to classify tumors in images. A breast
cancer diagnosis system proposed in [17] was based on FL-GA to detect the
Wisconsin breast cancer diagnosis (WBCD). The system has been tested and show
average accuracy of 97%. In addition, [18] proposed a FL-GA approach for the
classification of epilepsy risk level from EEG signals. This technique performed well
with the accuracy reaching more than 90%. A system has been developed to predict
the lung sounds using a hybrid system of NN-GA [19]. Another application of NN-
GA was also developed to analyze digital mammograms [20]. A soft computing
methodology is demonstrated when NN, FL and GA were combined in one system to
solve a specific problem. A system that combined the three methodologies was
proposed in [21]. The purpose of the system was to interpret medical images by
classifying and detecting any abnormalities. Other systems in [22] and [23] utilized
the same methodology and the results of all systems have been reported to be very
good.
Methodologies such as NN-FL, NN-GA and FL-GA have been used in many
applications. However, in regard to the technique where the three methodologies are
combined in a hybrid system, there are still limited applications of this approach.
According to [24], NN-FL-GA has the smallest number of published papers. Only 5
papers in the period 1995-2007. NN-FL-GA methodology has a great advantage
either in complementing each other or in solving each others limitation. Hence, more
applications should be developed in computer aided diagnosis where the accuracy is
the most important issue.
Hybrid Intelligent System for Disease Diagnosis 131
Combination Description
NN-FL NN controlled by FL
FL controller tuned by NN
NN-GA GA used to pre-process the
input parameters that train the
NN
FL-GA GA controlled by FL
FL controller tuned by GA
Fig. 1. Activity diagram of the first phase (learning phase) of the system
Fig. 2. Activity diagram of the second phase (diagnosis phase) of the system
The architecture of the proposed system is in figure 3 which shows the integration
of the training phase and the diagnosing phase.
Hybrid Intelligent System for Disease Diagnosis 133
Fig. 3. The proposed architecture of a hybrid disease diagnosis system using neural network,
fuzzy logic and genetic algorithm
The acquired image(s) will be filtered by the system to reduce the noises acquired
during the acquisition process. In this project, Wavelet Transform (WT) is one of the
techniques that is used in signal processing to suppress noise. WT can be defined as
follows [25]:
(1)
(2)
Where the Pm is a signal that is the approximation of the f(x) at a resolution m, and the
Qm is a signal that is the difference between the approximation Pm 1 and Pm.
134 H.R.H. Al-Absi et al.
After the images have been de-noised, a feature extraction process will be applied to
extract the most important features for image classification. Texture features will be
extracted here to obtain descriptors that can lead to better performance. It has been
proven that texture descriptors are more effective than shape descriptors in providing
more information of the images [26]. PCA-Wavelet technique will be used to extract
the features and to segment the image(s). The feature extraction process is conducted
in which 2D-Discrete wavelet transform decomposes an image to reduce its resolution
and then the PCA uses eigenfeatures to extract object features from the image [27].
After the features extraction, the next step is to use these features as input to the
learning algorithm. The algorithm combines the strength of the three methodologies
to form a strong algorithm. The strengths include the optimization of GA, the learning
capability of ANN, and the uncertainty and imprecision tolerance of FL that will be
utilized in the learning of the ANN.
Since the system is expected to be able to retrain itself if it is not being able to
decide whether an image contains abnormalities or not, and since retraining the whole
network is time consuming and not practical, the network type that will be used in this
system must have some sort of memory that stores the output weights of the network
after training, which to be combined with the new data. Recurrent Neural Network
(RNN) [28] is a type of artificial neural networks that enclose a memory or a context
layer that memorizes / keeps the outputs weights of the hidden layer. The context
layer uses these weights as an input to the hidden layer after a period of time (small
delay in time) [28] this can be utilized to retrain the network once new data is
presented and a retraining of the system is needed.
In this hybrid system, the first step is selecting the most significant features out of a
huge number of features that were extracted. Genetic algorithm performs feature
selection in order to reduce the amount of features used to train the algorithm. In
addition to that, GA is used to optimize the initial weights of the neural network.
Although artificial neural networks capability in learning has made it applicable in
many applications, however, the network deals only with crisp values which make it
not suitable in application that are based on imprecision or uncertainty. For that, in
order to enhance the capability of ANN under uncertainty or partially truth situation,
it is combined with fuzzy logic to improve the performance. Fuzzy rules are
represented in if-then approach, e.g. they can be expressed in the form if x is A, then y
is B. The role of fuzzy here is to deal with uncertainty and imprecision and combine
those features with the learning ability of neural network. Therefore, by combining
these two methods in a medical imaging application (which has lots of uncertainties
due to the nature of images and information enclosed in them i.e. organs,
abnormalities...etc); the system would contribute to minimize errors and maximize the
diagnosis accuracy. On the other hand, in fuzzy logic, not all fuzzy rules would
contribute to the best results. Therefore, they need to be refined. For this, GA is used
to optimize the fuzzy rules by selecting the most significant rules making the
performance of the algorithm better. Figure 4 shows the learning algorithm using the
combined soft computing techniques of NN-FL-GA.
Hybrid Intelligent System for Disease Diagnosis 135
4 Experiments
For the purpose of evaluating the accuracy of the hybrid learning algorithm presented
in the preceding section, experiment was conducted using NeuroSolutions software to
test the algorithm. NeuroSolutions is a simulation environment that provides a
platform for the development of neural network applications.
Col1 Col2 Col3 Col4 Col5 Col6 Col7 Col8 Col9 Col10 Col11
63375 9 1 2 6 4 10 7 7 2 4
76389 10 4 7 2 2 8 6 1 1 4
95719 6 10 10 10 8 10 7 10 7 4
128059 1 1 1 1 2 5 5 1 1 2
142932 7 6 10 5 3 10 9 10 2 4
144888 8 10 10 8 5 10 7 8 1 4
145447 8 4 4 1 2 9 3 3 1 4
160296 5 8 8 10 5 10 8 10 3 4
167528 4 1 1 1 2 1 3 6 1 2
183913 1 2 2 1 2 1 1 1 1 2
136 H.R.H. Al-Absi et al.
The hybrid learning algorithm was built using the NeuroSolutions graphical interface
tool. All parts of the system were connected to form the learning algorithm for the
three methods experimented. Figure 5 illustrates the NeuroSolutions Breadboard of
the hybrid learning algorithm presented in this paper.
To evaluate the hybrid learning algorithm and comparing it with other methods,
699 samples were used in each experiment, where 60 % is for training, 20% is for
cross validation and 20% for testing (139 samples). Table 3 reports the results
obtained from the experiments.
As shown in the table, the performance of the proposed method is better than the
neural network with fuzzy logic, however, neural network with genetic algorithm is
better than the hybrid method. This could be due to the amount of the sample data that
was used for the training phase. Neural network applications require large dataset for
the training phase to produce better performance and with the combination of the
three methodologies in a hybrid system, the sample data for training should be larger
than the data that was used in this experiment.
Generally, the performance reported in these experiments shows some evidences
that the hybrid system may perform better than other methods which were discussed
in the previous sections.
Hybrid Intelligent System for Disease Diagnosis 137
No. of Testing
Method samples Correct Incorrect Accuracy
References
1. Rangayyan, R.M., Ayres, F.J., Desautels, J.E.L.: A review of computer-aided diagnosis
of breast cancer: Toward the detection of subtle signs. Journal of the Franklin Institute
344(3-4), 312348 (2007)
2. Chen, D.-R., Hsiao, Y.-H.: Computer-aided Diagnosis in Breast Ultrasound. J. Med.
Ultrasound 16(1), 4656 (2008)
3. Huang, Y.-L.: Computer-aided Diagnosis Using Neural Networks and Support Vector
Machines for Breast Ultrasonography. J. Med. Ultrasound 17(1), 1724 (2009)
138 H.R.H. Al-Absi et al.
4. Marcela, X.R., Pedro, H.B., Caetano, T.J., Paulo, M.A.M., Natalia, A.R., Agma, J.M.:
Supporting content-based image retrieval and computer-aided diagnosis systems with
association rule-based techniques. Data and Knowledge Engineering 68(12), 13701382 (2009)
5. Nishikawa, R.M.: Current status and future directions of computeraided diagnosis in
mammography. Computerized Medical Imaging and Graphics 31, 224235 (2007)
6. Nishikawa, R.M.: Current status and future directions of computer-aided diagnosis in
mammography. Computerized Medical Imaging and Graphics 31(4-5), 224235 (2007)
7. Salas-Gonzalez, D., Grriz, J.M., Ramrez, J., Lpez, M., lvarez, I., Segovia, F., et al.:
Computer aided diagnosis of Alzheimers disease using support vector machines and
classification trees. Phys. Med. Biol. 55, 28072817 (2010)
8. Goggin, L.S., Eikelboom, R.H., Atlas, M.D.: Clinical decision support systems and
computeraided diagnosis in otology. OtolaryngologyHead and Neck Surgery 136, S21
S26 (2007)
9. Economou, G.-P.K., Lymberopoulos, D., Karvatselou, E., Chassomeris, C.: A new concept
toward computer-aided medical diagnosisA prototype implementation addressing
pulmonary diseases. IEEE Trans. Inform. Technol. Biomed. 5, 5566 (2001)
10. Stoitsis, J., Valavanis, I., Mougiakakou, S.G., Golemati, S., Nikita, A., Nikita, K.S.:
Computer aided diagnosis based on medical image processing and artificial intelligence
methods. Nuclear Instruments and Methods in Physics Research 569, 591595 (2006)
11. Ramirez, J., Chaves, R., Grriz, J.M., Lpez, M., lvarez, I., Salas-Gonzalez, D., Segovia,
F., Padilla, P.: Computer aided diagnosis of the Alzheimers Disease combining SPECT-
based feature selection and Random forest classifiers. In: IEEE Nuclear Science
Symposium Conference Record (NSS/MIC), Orlando, FL, pp. 27382742 (2009)
12. Azvine, B., Azarmi, N., Tsui, K.C.: An introduction to soft computing A tool for
building intelligent systems. LNCS, pp. 191210. Springer, Heidelberg (1997)
13. Negnevitsky, M.: Artificial intelligence: A Guide to Intelligent Systems, 2nd edn. Addison
Wesley (2005)
14. Nazmy, T.M., El-messiry, H., Al-bokhity, B.: Adaptive neuro-fuzzy inference system for
classification of ecg signals. Journal of Theoretical and Applied Information Technology
(2005)
15. Raja, K.B., Madheswaran, M., Thyagarajah, K.: A Hybrid Fuzzy-Neural System for
Computer-Aided Diagnosis of Ultrasound Kidney Images Using Prominent Features.
Journal of Medical Systems 32 (2008)
16. Benamrane, N., Freville, A., Nekkache, R.: A Hybrid Fuzzy Neural Networks for the
Detection of Tumors in Medical Images. American Journal of Applied Sciences 2(4), 892
896 (2005)
17. Andre, C., Pen-Reyes, Sipper, M.: A fuzzy-genetic approach to breast cancer diagnosis.
Artificial Intelligence in Medicine 17, 131155 (1999)
18. Harikumar, R., Sukanesh, R., Bharathi, P.A.: Genetic algorithm optimization of fuzzy
outputs for classification of epilepsy risk levels from EEG signals. In: Conference Record
of the Thirty-Eighth Asilomar Conference on Signals, System and Computer (2004)
19. Guler, I., Polat, H.u., un, U.m.E.: Combining Neural Network and Genetic Algorithm for
Prediction of Lung Sounds. Journal of Medical Systems 29(3) (2005)
20. Verma, B., Zhang, P.: A novel neural-genetic algorithm to find the most significant
combination of features in digital mammograms. Applied Soft Computing 7, 612625
(2007)
21. Benamrane, N., Aribi, A., Kraoula, L.: Fuzzy Neural Networks and Genetic Algorithms for
Medical Images Interpretation. In: Proceedings of the Geometric Modeling and Imaging
New Trends (2006)
Hybrid Intelligent System for Disease Diagnosis 139
22. Ozekes, S., Osman, O., Ucan, O.N.: Nodule Detection in a Lung Region thats Segmented
with Using Genetic Cellular Neural Networks and 3D Template Matching with Fuzzy Rule
Based Thresholding. Korean Journal of Radiology (2008)
23. Das, A., Bhattacharya, M.: GA Based Neuro Fuzzy Techniques for Breast Cancer
Identification. In: International Machine Vision and Image Processing Conference (2008)
24. Yardimci, A.: Soft computing in medicine. Applied Soft Computing 9, 10291043 (2009)
25. Satish Chandra, D.V.: Image Enhancement and Noise Reduction Using Wavelet
Transform. In: Proceedings of the 40th Midwest Symposium on Circuits and Systems
(1997)
26. Muhammad, M.N., Raicu, D.S., Furst, J.D., Varutbangkul, E.: Texture versus Shape
Analysis for Lung Nodule Similarity in Computed Tomography Studies. In: Andriole,
K.P., Siddiqui, K.M. (eds.) Medical Imaging 2008: PACS and Imaging Informatics.
Proceedings of the SPIE, vol. 6919 (2008)
27. Kim, K.-A., Oh, S.-Y., Choi, H.-C.: Facial feature extraction using PCA and wavelet
multi-resolution images. In: Sixth IEEE International Conference on Automatic Face and
Gesture Recognition (2004)
28. Elman, J.L.: Finding structure in time. Cognitive Sci. 14(2), 179211 (1990)
29. Mangasarian, O.L., Wolberg, W.H.: Cancer diagnosis via linear programming. SIAM
News 23(5), 118 (1990)