結論與建議

當病患和醫師還未產生醫病關係時，且病患也不曉得自己身體所 呈現的疼痛病症應至哪一科就診時，就算病患到服務櫃檯或由掛號人 員現場處理就診科別，通常造成誤導病患重複就醫，浪費醫療資源。

而在一般的網路掛號系統中，更無掛號諮詢服務。本研究依遵循系統 設計原則以倒傳遞網路建立網路掛號作業。利用專家的知識建立掛號 諮詢知識庫，且利用倒傳遞網路（BPN）來預測分析，如經由倒傳遞 網路所分析的資料，必透過專家的修改，才能納入訓練資料表中，以 便利回想作業。，以玆減少醫療資源浪費。其中所得的結論如下所示：

1. 本研究提供病患較準確的就診法則，因可利用倒傳遞網路讀取專 家知識庫資料來分析預測，建議病患至適當的科別就診。

2. 類神經網路掛號系統的學習準確度很高，可處理複雜的樣本識別 問題。因倒傳遞演算法收斂速度很慢，所以學習速度也很慢，但 因系統會將此作業安排至離峰時間執行。至於什麼樣的收斂結果 是最好的呢？其答案是不曉得，但盡量收斂於一局部最小誤差 值。

3. 類神經網路掛號系統的回想速度快，且訓練資料表是由法則庫轉 換而成的，所以學習新樣本時也不會遺忘已學習過樣本。

4. 當病患所要輸入的病症，必須在訓練資料表中有資料才行，也就

是如要提高病症與就診科別的相關程度，即必須病症對應就診科 別法則中要有多筆的病症對應就診科別之交集關係。

後續研究可依倒傳遞網路演算法將此系統發展成家庭醫學專家

系統，只需將檢驗檢查的病症名稱與數據納入系統中，依照實證醫學

的疾病準則去分析出為何種疾病。

參考文獻

英文部分

1. Novak, B. “Superfast autoconfiguring artificial neural networks and their

application to power systems”, Electric Power Systems Research, Vol.35 (1),

Oct., 1995, pp. 11-16.

2. Chrwan-Jyh Ho, and Hon-Shiang Lau “Evaluating the impact of operating

conditions on the performance of appointment scheduling rules in service system”, European Journal of Operational Research, Vol.112, 1999, pp.542-553.

3. Burr, D.J. “Experiments on neural net recognition of spoken and written text”, IEEE Trans. on ASSP., Vol.36(7), 1997, pp.1162-1168.

4. Lin, J. S., Cheng, K. S., and Mao, C.W. “Multispectral magnetic resonance

images segmentation using fuzzy Hopfield neural network”, Inter. J. of

Bio-Medical Computing, Vol.42 (3), Aug., 1996, pp.205-214.

5. Kosko, B. “Bidirectional associative memories”, IEEE Trans. on System, Man.

and Cybernetics, 1987.

6. Pregenzer, M., Pfurtscheller, G. and Flotzinger, D. “Automated feature selection

with a distinction sensitive learning vector quantizer”, Neurocomputing, Vol.11

(1), May, 1996, pp. 19-29.

7. Nelson, M.M., and Illingworth, W.T. “A practical guide to neural nets”, Addison-Wesley, 1990.

8. Mehrotra, Kishan, Chilukuri K. Mohan, and Sanjay Ranka, “Elements of

Artificial Neural Networks”, London：A Brodford Book The MIT Press, 1997,

pp.8.

9. Mei-Chiao Shen, “A Simulation on an Internist Office Appointment System”, Journal of the Overseas Chinese College of Commerce, Vol.11, 1993, pp.103-114.

10. Patrick Wang, P. ”Sequencing and scheduling N customers for a stochastic

server”, European Journal of Operational Research , Vol.119,1999, pp.729-738.

11. Tigges P., Norbert, K. and et al. “Identification of input variables for feature

based artificial neural networks-saccade detection in EOG recordings”, Inter. J.

of Medical Informatics, Vol.45 (3), Jul., 1997, pp. 175-184.

12. Zhang, P. and Chen, L. “A novel feature extraction method and hybrid tree

classification for handwritten numeral recognition”, Pattern Recognition Letters,

Vol.23, 2002, pp. 45-56.

13. Amari, S. “Mathematical foundation of neurocomputing”, Proc. IEEE, Vol.78 (9), 1990, pp.1443-1463.

14. Fukuda, T., and Shibata, T. “Theory and Applications of Neural Networks for

Industrial Control Systems”, IEEE Trans. on IE, Vol.39(6), 1992, pp.472-489.

中文部分

附錄 A：METMBS'02 International Conference

The 2002 International Conference on Mathematics and Engineering Techniques in Medicine and Biological Sciences

An Intelligent Appointment System Based on the Neural Networks

Fan Wu and Han-Ping Sun Institute of Health Medical Service

China Medical College, Taiwan

Email: [email protected], Tel: 886-4-24373676, Fax: 886-943272049.

Abstract

For avoiding getting in lines, most hospitals provide convenient ways for patients to make an appointment easily. A web-based appointment system can facilitate the patients to get the related information about the service time of each doctor and to make an appointment.

However, the medical knowledge of the patients is limited. They can not always make a correct decision about which department they should visit when they suffer the uncomfortable.

This paper establishes an on-line consulting system. It gets the symptoms and the concerns from the patients and then provides them the advices about which department the patients should visit. The system contains the knowledge base and rule base. The system first tries to match the input symptoms with the contents in the knowledge base. If the input symptoms do not exactly match data in the knowledge base, a backward-propagation neural network will be used to perform the classification of the possible departments. The system can suggest the patients the appropriate departments according to their symptoms. With the advices, the patient has higher satisfaction and better outcomes.

Keywords: Appointment system, symptom, neural network, hospital information system, closed staff hospital.

1. Introduction

The hospitals run in the style of closed staff hospitals do not accept new applications from the doctors that are not employed by the hospitals for any category

of medical service except contracts. In such a condition, patients will make an appointment by themselves through the appointment scheduling unit of the hospital before they visit the hospital. The scheduling unit will offer the next available time slot for the patients. For avoiding getting in lines, the hospitals generally provide the convenient ways such as internet and the kiosks such that patients can make an appointment easily.

Establishing a web-based appointment scheduling system can help patients to easily get the related information about the service time of each doctor. The patients can see the disease lists of various departments that the departments can take care.

The patients can select which department they should visit the hospital according to their symptoms. Since there is no medical staff or expert on internet to provide consulting service, the patient sometimes may make an appointment to visit the inappropriate department. Since the medical knowledge of the patient is limited, the patients can not correctly decide which department they should visit. For example, it is not an easy job for a woman to decide which department (digestion internal medicine or obstetrics and gynecology) she will make an appointment with the clinician when she got abdominal pain.

The financial condition of the health insurance is worse in each country, the patient does not visit the appropriate department will not only delay the medical treatment but also waste the medical resource. This paper will establish an on-line consulting system to provide the patient the related medical knowledge. The goal of the system is to aid the patients in making the appointment, supplying the correct department in relation to the symptom of the patient. The system gets the symptoms and the concerns of the patients and then provides the advices about which department the patients should visit. For making a valuable advice, the system has a database containing the information of the verified consultations. In case the symptoms of a patient exactly match the data in the database, the verified consultations are provided directly. If the symptoms of the patient are partially match the data in the database, the system will use the neural network to infer which the department that the patient should be taken cared. After a period of testing, the reward from the physicians and the patients is positive. The system can reduce the patients’

anxiety and fear and increase in peace of mind. In addition, the patient has higher satisfaction and better outcomes.

2. Background

In an open staff hospital, the patient is arranged to be hospitalized by his family physician. The family physician books the appointment for the clinical services of the hospital to treat the patient. Under this condition, the appointment system of the

hospital is a scheduling system whose major function is to allocate and arrange time and space to furnish certain clinical service [1, 2]. There is a lot of research about the arrangement of the appointment system. The system can simplify the scheduling requirement such as on call schedule, after-hours appointment, emergencies and surgeries.

In a closed staff hospital, its outpatient service allows all patients to come to see a doctor with/without the referring from the family physician. The patients can make an appointment by themselves through internet. However, there is little literature talking about the appointment system that works as expert and provides the consultation to patients about which department they should visit. The Artificial Neural Network (ANN) is an information-processing paradigm that emulates the properties of biological nervous systems and draws on the analogies of adaptive biological learning [3, 4]. Learning in biological systems involves adjustments to the synaptic connections that exist between the neurons. This is true of ANN as well [5].

Learning typically occurs through training, or exposures to a trusted set of input/output data where the training algorithm iteratively adjusts the connection weights. The ANNs are good pattern recognition engines and robust classifiers with the ability to generalize in making decisions about imprecise input data. They offer ideal solutions to a variety of classification problems such as speech [6], character [7]

and signal recognition [8], as well as functional prediction and system modeling where the physical processes are not understood or are highly complex. The advantage of ANNs lies in their resilience against distortions in the input data and their capability of learning.

There are multitudes of different types of ANNs. Some of the more popular include the multilayer perception which is generally trained with the back-propagation of error algorithm [3, 9, 10], learning vector quantization [3, 11], Hopfield [12], and Kohonen [13], to name a few. A way of classifying ANN types is by their method of learning (or training), as some ANNs employ supervised training while others are referred to as unsupervised or self-organizing [4, 5]. Supervised training is analogous to a student guided by an instructor. Unsupervised algorithms essentially perform clustering of the data into similar groups based on the measured attributes or features serving as inputs to the algorithms. This is analogous to a student who derives the lesson totally on his or her own.

3. Intelligent Appointment System 3.1 The structure of appointment system

Making an appointment through the internet access is a convenient way for patients to reduce the waiting time. Since no on-line expert exists to provide the

consultation, it often results in repetition cure to the patient because the inappropriate department. For improving the drawback, we develop an expert system as the appointment consultant on the Web. The system is constituted by knowledge base and rule base. An inference engine is designed to operate upon the above two bases. The system has a gateway that can propagate the result of the consultant to the Hospital Information System (HIS). The structure of the system is shown in Figure 1.

Figure 1. The structure of the expert system for the appointment system.

3.2 The components of system

The patients may suffer the uncomfortable with many symptoms. For simplifying the expert system, five types of symptoms at most are allowed to input the system. The system will search the knowledge base. If the input symptoms exactly match the symptoms in the knowledge base, the corresponding department will be provided. Otherwise, the expert system will invoke the ANN to classify the department according to the symptoms. The detail discussion of the component is in the following.

3.2.1 Knowledge Base

The contents of the knowledge base are supplied and maintained by the USER or EXPERT

physicians or experienced medical staffs. The knowledge base stores the symptoms and the suggested department in correspondence to the symptoms. A relational database is used to develop the knowledge base, in which a table stores the contents of the knowledge. Each row in the table contains seven fields, as is shown in Figure 2:

the first five fields from smp1 to smp5 store the symptoms, the field dptmnt stores the suggested department and the field exp stores the explicitly tagged condition of the suggested departments (discussed later). When the patient inputs the symptoms through the web, the system will compare them with the contents in the table. If the fields from smp1 to smp5 of some row in the table exactly match the input symptoms, the related departments will be provided to the patient. Since the same set of symptoms can be treated by the different departments, the system may have more than one row with the same symptoms but with the different departments.

Create table sym_to_dep (

simp1 char(20),

Figure 2. The scheme of the table containing the knowledge base.

Assume the system provides fifty types of symptoms that the patients can select.

Theoretically, the set of symptoms will have up to 2⁵⁰ combinations of conditions (without permutation); that is, there will be 2⁵⁰ rows of data in the database. Most patients will specially care several uncomfortable symptoms. Thus the system limits the number of symptoms the patients can input is five at most. Thus the number of the rows in the database will at most be

×

50

C1 C₁⁴⁹× C₁⁴⁸×C₁⁴⁷× C₁⁴⁶ (≅ 2×10⁸) (1) The number in Equ.(1) is very big. We try to analyze the medical domain to reduce the number. We know that some symptom is certainly to be treated by specific department. For example, if some patient has the symptom of teeth aching, he must visit the dentist; no matter he has other symptoms. Assume a patient has the symptoms of teeth aching but also other symptoms. The system will response the dental department according to the teeth aching and then infers out other department(s)

from the remaining symptoms. For this condition, the system has a tag, i.e., field exp, to explicitly mark the symptom that will be isolated from the other symptoms for inferring the department(s). To be detail, let set S is composed by such the symptoms.

When the patient inputs the set of symptoms (let it denote the symptom set T) to the system, the system will first check whether the set of T intersects the set of S. Assume the answer is positive (i.e.,

T I S

≠

φ

), and P is the set of the intersection (i.e., P =

T I ). The system will first select the related departments corresponding to P. Then S

the system excludes the symptoms in set P from set T and continues the remaining comparison for other possible departments.

For the sake of simplicity, we restrict each row in the table contains one symptom at most to deal with the above condition. Assume there are k symptoms such that only one symptom of them can determine the department. The largest number of the rows in the table will be

×

C

₁⁵⁰−k

C

₁^49−k×

C

₁^48−k ×

C

₁^47−k×

C

₁^46−k (2) If the value of k is 5, the value of Equ.(2) will be about 1×10⁶, which reduces to about 1/200 of the value in Equ.(1). Though the number of the combination reduces a lot, it is hard to exhaust to enumerate all combination of the symptoms.

The medical staffs may input several data or rules. The system will use the rules to generate the combination of the symptoms and the related department(s) into the knowledge base. If the symptoms of the patient still do not appear in the knowledge base, the system will invoke the ANN to classify the department.

3.2.2 Rule Base and Inference Engine

Medical diagnosis is one of the earliest domains approached with the rule-base system. For the appointment rule, we have interviewed experienced medical staffs and from these interviews collected rules about the relationships between the symptoms and the departments. We capture the expert’s expertise and encode this expertise as rules in a nested if-then-else format for classification. The if part is the premise of the rule, and the then part is the consequent. The else component of the consequent is optional.

The inference engine will generate the appointment knowledge based upon the rule. The reasoning mechanism used by the inference engine to process rules is simple.

The premise of the rule will be stored in the fields from symp1 to symp5, the consequent value of the rule is assigned in the filed of dptmnt. For example, if the scenario of the rule is:

if symptom A then if symptom B then output department α else output department β.

Then inference engine will output two rows into the knowledge base in the following:

symp1 symp2 symp3 symp4 symp5 dptment exp

A null Null null null α null

A B Null null null β null

3.2.3 Classification by the Neural Network

The system will first try to match the input symptoms with the contents in the knowledge base. If the input symptoms do not exactly match the data in the knowledge base, the system will manage the classification of the possible departments by a backward-propagation neural network.

The network has two layers of neurons (nodes): the first layer contains five nodes respectively representing the input nodes from symptom 1 to symptom 5; the second layer contains fourteen nodes representing the fourteen departments of the hospital.

Each node in the first layer is fully connected to the all nodes in the second layer, with each connection having a weight on it. The network works in two stages: one is for learning and the other for recalling. Learning involves adjustments to the weight of the connections existing between the nodes. The learning is in a supervised mode guided by contents of the knowledge base. The program takes the data in the knowledge base, computes the output based on the current connection weights and then compares the output with the desired. The training is performed iteratively until the adjustment of the connection weight converges. In general the converge will need large times of iterations, as is the properties of the backward propagation algorithm.

After the training process, the network can be used to manage the unseen pattern of symptoms. The usage of the neural network is under the condition that when the input symptoms do not exactly match the contents of the knowledge base. The recall function of the system is invoked to run the assessment process under the current connection weights. The opinions from the medical staffs say that the advices are near the ones provided by the experts.

The neural network continues learning off line to increase the accurate of the next time. The training process will redo when new data are inserted into the knowledge base and confirmed by the experts.

4. Conclusion

An intelligent appointment system able to provide the advices has been developed. The system uses the data base to store the knowledge. The rule base provides a clean and easy interface to get the expertise from the experienced staff. The inference engine is executed beforehand to transfer the rules to the knowledge base.

With this way, the recall process can be shortened by the high processing ability of the relational database. The neural network is used to classify when the knowledge base

does not match the input condition. The network can adaptively learn when the new confirmed data enter the knowledge base. In agreement with the present symptoms, the system is show to be able to get the appropriate advices to the patients.

REFERENCE

1. Wang, P.P. “Sequencing and scheduling N customers for a stochastic server”, European J. of Operational Research ,Vol. 119, 1999, pp.729-738.

2. Ho, C.J. and Lau, H.S. “Evaluating the impact of operating conditions on the

performance of appointment scheduling rules in service system”, European J. of

Operational Research ,Vol.112, 1999, pp.542-553.

3. Hecht-Nielsen, R. “Neurocomputing”. Addison-wesley, Reading, MA.

4. Amari, S. “Mathematical foundation of neurocomputing”, Proc. IEEE, Vol.78 (9), 1990, pp.1443-1463.

5. Rupali, P.V., Tambe, S.S., and Kulkarni, B.D. “ANN modeling of DNA sequences:

new strategies using DNA shape code”, Computers and Chemistry, Vol.24 (6),

Sep., 2000, pp. 699 – 711.

6. Burr, D.J. “Experiments on neural net recognition of spoken and written text”, IEEE Trans. on ASSP., vol.ASSP-36, No.7, 1997, pp.1162-1168.

7. Zhang, P. and Chen, L. “A novel feature extraction method and hybrid tree

classification for handwritten numeral recognition”, Pattern Recognition Letters,

Vol. 23, 2002, pp. 45-56.

8. Tigges, P., Norbert, K., and et al. “Identification of input variables for feature based

artificial neural networks-saccade detection in EOG recordings”, Inter. J. of

Medical Informatics, Vol.45 (3), Jul., 1997, pp. 175-184.

9. Nelson, M.M. and Illingworth, W.T. “A practical guide to neural nets”, Addison-Wesley, 1990.

10. Novak B. “Superfast autoconfiguring artificial neural networks and their

application to power systems”, Electric Power Systems Research, Vol.35 (1), Oct.,

1995, pp. 11-16.

11. Pregenzer, M., Pfurtscheller, G., and Flotzinger, D. “Automated feature selection

with a distinction sensitive learning vector quantizer”, Neurocomputing, Vol.11

(1), May, 1996, pp. 19-29.

12. Lin, J. S., Cheng, K. S., and Mao, C.W. “Multispectral magnetic resonance

images segmentation using fuzzy Hopfield neural network”, Inter. J. of

Bio-Medical Computing, Vol.42 (3), Aug., 1996, pp.205-214.

13. Song, X.H. and Hopke, P.K. “Kohonen neural network as a pattern recognition

method based on the weight interpretation”, Analytical Chemical Acta, Vol.334,

Nov., 1996, pp.57-66.

附錄 B：2001 電子商務理論與實務研討會

應用，以網路之技術來開發構建一套醫院轉診作業管理系統，此系統的優點可以幫助原診治醫療

四、系統之建構

3. 原診治醫療機構的醫師可透過網路來查詢病患最新的治療狀況與病患資料。

圖 2. 轉診網路作業流程圖

本研究構建之轉診作業系統，以上述之系統需求分析為出發點，則原診治醫療機構可以透過 該系統進行轉診輸入、查詢作業，而接受轉診醫療機構可以透過該系統進行轉診處理、列印作業，

其系統主要的構成包括轉診資料庫與 WEB 程式介面，其系統架構如圖 3 所示，其說明如下：

一、轉診資料庫

儲存轉診所需的資料，其主要的資料內容為保險對象基本資料與病歷摘要等，其中保險對象 基本資料的姓名、身分證號、出生日期為必要的輸入資料。目前是以 MS SQL 資料庫架設。

儲存轉診所需的資料，其主要的資料內容為保險對象基本資料與病歷摘要等，其中保險對象 基本資料的姓名、身分證號、出生日期為必要的輸入資料。目前是以 MS SQL 資料庫架設。

在文檔中 類神經網路為基礎之智慧型醫院網路掛號系統; An Intelligent Appointment System Based on the Neural Networks (頁 72-92)