This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
The application of machine learning tools has shown its advantage in medical aided decision. The purpose of this study is to construct a medical decision support system based on support vector machines (SVM) with 30 physical features for helping the Doctors Specialized in Anesthesia DSA in pre-anesthetic examination or preoperative consultation. For that, in this work, a new dataset has been obtained with the help DSA. The patients (898 patients) in this database were selected from different private clinics and hospitals of western Algeria.
The medical records collected from patients suffering from a variety of diseases ensure the generalization performance of the decision system.
In this paper, the proposed system is composed of four parts where each one gives a different output. The first step is devoted to automatic detection of some typical features corresponding to the ASA (American Society of Anesthesiologists) sores. These characteristic are widely used by all DSA in pre-anesthetic examinations. In the second step, a decision making process is applied in order to accept or refuse the patient for surgery. The goal of the following step is to choose the best anesthetic technique for the patient (general anesthesia or local anesthesia). In the final step we examines if the patient's tracheal intubation is easy or hard.
Moreover, the robustness of the proposed system were examined using 6-fold cross-validation method and results show the SVM-based decision support system can achieve an average classification accuracy of 87.52% in the first module, 91.42% for the second module, 93.31% for the third module and finally 94.76 % for the fourth module.
Keywords- Doctors Specialized in Anesthesia, Support vector machines, American Society of Anesthesiologists scores, machine learning, pre-anesthetic examination.
Manuscript received "Date here here"
About 1st author: Biomedical engineering laboratory, Tlemcen University
(Telephone: +213Â 550Â 568 090 email: email@example.com)
About 2nd author: Biomedical engineering laboratory, Tlemcen University, System and modeling research unit, Liege university
( email: firstname.lastname@example.org)
About 3rd author: Biomedical engineering laboratory, Tlemcen University
About 4th author: Biomedical engineering laboratory, director of CREDOM research unit, Tlemcen University
The risks of anesthesia and mortality rates are pretty low these years. As a matter of fact, not only have errors become relatively uncommon, but experts say that anesthesia is one of the safest areas of health care today thanks to the works being done in the medical decision support in the field of anesthesia.
In the word, medical students are few to move towards the profession Doctors Specialized in Anesthesia DSA, the number of DSA tends to decrease very disturbing. In Algeria there are about 7,000 DSA, a number which is insufficient to insure all the tasks that have to be performed for the safety of the patients.  The main problem is that despite their small number, their presence is indispensable in each hospital or clinic. Indeed, they have to insure the pre-anesthetic examinations of all patients who need general or local anesthesia. Moreover, they have to be present in the operating room during surgery and after that during the hospitalization (post-operative period).
The realization of these different tasks is really hard to perform. That is why, we propose in this work an artificial intelligence based approach allowing to bring assistance to the DSA.
The related works in preoperative patient classification was carried out by Peter et al. in . The authors have developed an automatic instrument used for grading the level of anesthetic patient risk, with a modified version presented by Hussman and Russell . So far, risk prediction has been carried out using statistical analysis tools, which lacks the desired precision .
In the same field, another work has been done in . Authors propose a Support Vector Machine -SVM- based decision system for clinical aided tracheal intubation and predication with multiple features. The experiments use 264 medical records and only one technique of classification. In this research, 30 basic and anthropometrical features in total were taken into consideration for the 898 patients.
Support vector machine (SVM) was applied to build an aided decision support system to estimate in the first step the ASA physical status. In the second step, a decision making process is applied in order to accept or refuse the patient for surgery. The goal of the following step is to choose the best anesthetic technique for the patient (general or local anesthesia). The final step examines if the patient's tracheal intubation is easy or hard. Furthermore, 6-fold cross-validation method was used to test the robustness of the proposed system and results showed that the SVM-based decision support system with 30 features could achieve high classification accuracy in each step
In this paper, we target two distinct objectives: the database construction and data classification. To this aim, we divide this work as follows. In section II, we describe the database used and we discuss its different parameters. After that in section III, reviews some basic SVM concepts. Section IV presents the experimental results and discussion. Finally, we shall summarize the main points of our prototype and conclude the paper.
In this section we present the creation of the dataset. The dataset has been obtained with the help of DSA. The patients in this database were selected from different private clinics and hospitals of western Algeria (TLEMCEN hospital, ORAN CANASTEL hospital, ORAN HAMMOU BOUTELILIS clinical, ORAN NOUR clinical, TLEMCEN LAZOUNI clinical).
We have to notice that the unavailability of a standardized database in this field forced us create these personal database. In total 898 subjects participated in the data collection, 488 males and 410 females.
Our database is divided into four sub-bases. Each sub-base has a specific task to achieve. The first sub-base (SB1) is devoted to the detection of the ASA physical status. It is characterized by 17 parameters presented in table1.
488 males and 410 females
between 2 months and 105 years
Heart failure (HF)
Heart rate 1 (bpm)
Heart rate 2 (bpm)
Heart rate3 (bpm)
Steadiness of heart rate
Left ventricular hypertrophy
Take a measure (%)
Blood sugar or blood glucose level
Take a measure (g/l)
Blood pressure (mmHg)
Physical Status according to the examination by the DSA
Table.1. SB1 dataset parameters
The ASA physical status allows to evaluate the anesthetic risk and to obtain a predictive parameter of surgical mortality. We have selected patients with ASA Physical Status 1, 2, 3 and 4. We could not select patients with ASA Physical Status 5 and 6 because they were dying. The output (classes) for the database takes the values '1', '2', '3', or '4'. Statistics of ASA score in our data base is resumed in table2.
Where: '1' means a patient is in ASA physical status 1.
'2' means a patient is in ASA physical status 2.
'3' means a patient is in ASA physical status 3.
'4' means a patient is in ASA physical status 4.
They are 219 patients (24.38%) cases in class '1', 395 patients (43.98%) cases in class '2', 232 patients (25.84%) cases in class '3', and just 52 patients (05.80%) cases in class '4'.
The ASA physical status classification system is a system for assessing the fitness of patients before surgery.
In 1963 the American Society of Anesthesiologists (ASA) adopted the five-category physical status classification system. A sixth category was later added. These characteristics are presented in table3 .
ASA Physical Status
Number of patients
Mean age (year)
Mean heart rate 1 (bpm)
Mean heart rate 2 (bpm)
Mean heart rate 3 (bpm)
Mean oxygen saturation
Mean blood glucose level
Mean blood pressure (systole)
Mean blood pressure (diastole)
Table.2.Â Clinical data of all 898 patients and their distribution according to ASA class
ASA Physical Status 1
A normal healthy patient
ASA Physical Status 2
A patient with mild systemic disease
ASA Physical Status 3
A patient with severe systemic disease
ASA Physical Status 4
A patient with severe systemic disease that is a constant threat to life
ASA Physical Status 5
A moribund patient who is not expected to survive without the operation
ASA Physical Status 6
A declared brain-dead patient whose organs are being removed for donor purposes
Table.3. ASA Physical Status
The second sub-base (SB2), which is characterized by three attributes: the first one is the result of the first classifier (ASA Physical Status), the second is the cerebrovascular accidentÂ (CVA) and the third one being the myocardial infarction (MI).
These three parameters are exposed in table4. It aims at detecting if the patients are accepted or refused for
ASA physical status
The output of the first classifier
ASA1, ASA2, ASA3, ASA4
CerebroVascular Accident (CVA)
The CVA is a very serious condition in which the brain is not receiving enough oxygen (o2) to function properly. Cerebrovascular accidents are the second leading cause of death worldwide
If the duration of the deficit was <24 h, it was defined as a transient ischemic attack.
If the deficit persisted for a longer period, it was defined as a stroke. 
Myocardial InfarctionÂ (MI)
The Myocardial Infarction (MI) or acute myocardial infarction (AMI), commonly known as a heart attack, results from the interruption of blood supply to a part of the heart, causing heart cells to die. 
Accept patient for surgery
Refuse patient for surgery
Table.4. SB2 dataset parameters
If the patient has been subject to an MI and/or a recent CVA (less than 6 months), he is automatically refused or his surgery is put off to a later date.
As far as the first parameter of the second classifier is concerned, the ASA Physical Status can have the score 1, 2, 3 or 4 according to the physical status of the patient.
Concerning the second and third parameters, they are classified into three categories:
Category 0: is for patients who have never been subject to any CVA and/or MI
Category 1: is for patients who have been subject either to an CVA and/or an MI at least 6 months ago.
Category 2: is for patients who have been subject either to an CVA and/or an MI less than 6 months ago.
The output (classes) for SB2 takes the values '0', '1'.
Where: '0' means a patient is refused for surgery and '1' means a patient is accepted for surgery.
They are 136 patients (15%) cases in class '0', and 762 patients (85%) cases in class '1'.
The third sub-base (SB3) is devoted to the detection of the best anesthetic technique for the patient (general anesthesia or local anesthesia). It is characterized by three attributes: the first one is age, the second is the state of patients, the third is the body mass indexÂ (BMI), and finally types of surgery. These four parameters are exposed in table5.
Newborn, Child, Young, Adult, Old
State of patient
Normal, Mental illness, Hyper stressed, Down syndrome
Types of surgery
They are 25 types of surgery
Boddy Mass Index (BMI) (kg/m2)
BMI =A person's weight / height squared
Table.5. SB3 dataset parameters
The output (classes) for SB3 takes the values '0', '1'.
Where: '0' means a technique of surgery for patient is General anesthesia.
'1' means a technique of surgery for patient is General anesthesia.
They are 198 patients (22%) cases in class '0', and 700 patients (78%) cases in class '1'.
The fourth part of our work deals with a fourth classifier. It aims at detecting if the patient's tracheal intubation is easy or hard. The learning of this classifier is done by Sub-based 4 (SB4) which is characterized by five features: these parameters are exposed in table6.
1, 2, 3, 4
Distance between thyroid cartilage and menton
Backgrounds of hard tracheal intubation
Yes or no
Normal, Toothless, Upper dentures, Lower dentures, Brace
Easy tracheal intubation
Hard tracheal intubation
Table.6. SB4 dataset parameters
The output for SB takes the values '0', '1' where: '0' means a patient's tracheal intubation is easy and '1' means a patient's tracheal intubation is hard.
They are 700 patients (78%) cases in class '0', and 198 patients (22%) cases in class '1'.
As we have seen previously the database has been divided into four sub-bases (SB1, SB2, SB3 and SB4). In this work we manage 10 classes and 30 features as shown in table 7.
SB1: ASA Physical Status
SB2: Accept or refuse patient for surgery
SB3: General or local anesthesia
SB4: Easy or hard tracheal intubation
Table.7. Recapitulative of database
Fig. 1. Recapitulative of database histogram
In this section we present the proposed prototype (figure2), and a basic concepts of SVM classifier This process allows to classify patient according to ASA score, to accept or refuse patient for surgery, to choose the best
anesthetic technique for the patient (general anesthesia or local anesthesia), and also to evaluate if the patient's tracheal intubation is easy or hard.
Fig.2. Functioning of the prototype
Our prototype is divided into four parts as shown in figure 2, each of them uses an sub-based dataset (SB1, SB2, SB3, and SB4) as shown in the previous section. These ones were used for learning and test with SVM technique.
The first part is devoted to the detection of the ASA physical status by SB1 dataset. The second part uses SB2 dataset his role is to choose if the patients are accepted or refused for surgery. The third part is devoted to the detection of the best anesthetic techniques (general or local anesthesia) by SB3 dataset. And finally the fourth part work with SB4 dataset, its objective is to determine if the patient's tracheal intubation is easy or hard
Each part contains three units. The first is the dataset (SB1, SB2, SB3, and SB4), the second is training/test based module with SVM classifier (SVM module1, SVM module2, SVM module3, and SVM module4) and finally the results module.
Basic concepts of SVM classifier
Support vector machinesÂ (SVMs, alsoÂ support vector networks  areÂ supervised learningÂ models with associated learningÂ algorithmsÂ that analyze data and recognize patterns, used forÂ classificationÂ andÂ regression analysis. The SVM algorithm is based on the statistical learning theory
VapnikÂ and the current standard incarnation (soft margin) were proposed by Vapnik andÂ Corinna CortesÂ in 1995. 
More formally, a support vector machine constructs a hyperplane or set of hyperplanes in a high- or infinite-dimensional space, which can be used for classification, regression, or other tasks. Intuitively, a good separation is achieved by the hyperplane that has the largest distance to the nearest training data point of any class (so-called functional margin), since in general the larger the margin the lower the generalization error of the classifier.
Classifying dataÂ is a common task inÂ machine learning. Suppose some given data points each belong to one of two classes, and the goal is to decide which class aÂ newÂ data point will be in. In the case of support vector machines, a data point is viewed as aÂ p-dimensional vector (a list ofÂ pÂ numbers), and we want to know whether we can separate such points with a (pÂ âˆ’Â 1)-dimensionalÂ hyperplane. This is called aÂ linear classifier (as shown in Fig 3). There are many hyperplanes that might classify the data. One reasonable choice as the best hyperplane is the one that represents the largest separation, or margin, between the two classes. So we choose the hyperplane so that the distance from it to the nearest data point on each side is maximized. If such a hyperplane exists, it is known as theÂ maximum-margin (as shown in Fig 4)  hyperplaneÂ and the linear classifier it defines is known as aÂ maximumÂ margin classifier; or equivalently, theÂ perceptronÂ of optimal stability.
Fig.3. The sketch map of two class problem with SVM
Fig.4. Maximum-margin hyperplane and margins for an SVM trained with samples from two classes
It often happens that the sets to discriminate are not linearly separable in that space. For this reason, it was proposed that the original finite-dimensional space be mapped into a much higher-dimensional space, presumably making the separation easier in that space. To keep the computational load reasonable, the mappings used by SVM schemes are designed to ensure that dot products may be computed easily in terms of the variables in the original space, by defining them in terms of a kernel function K(x,y) selected to suit the problem (as shown in Fig 5).
Fig.5. Kernel machine for an SVM trained with samples from two classes
Experiments results and discussion:
The 6-fold cross-validation accuracy of each subset and mean accuracy are listed in Table8.
Each part of our prototype is presented as a confusion matrix (Table 8; 9; 10; 11). Usually, a confusion matrix contains information about actual and predicted classifications performed by a classification system. In this study, there are 10 diagnostic classes: in the first part, four classes (ASA physical status 1; 2; 3; and 4), in the second part, two classes (accepted or refused patient for surgery), in the third part, two classes (general or local anesthesia), and finally in fourth part, two classes (easy or hard patient's tracheal intubation).
In the confusion matrix, the rows represent the test data, while the columns represent the labels assigned by the classifier. Several indices of classification accuracy can be derived from the confusion matrix.
The cross-validation classification accuracy thus can be determined as:
part 1: (201+335+207+43) / 898 = 87.52%
part 2: (723+98) / 898 = 91.42%
part 3: (165+673) / 898 = 93.31%
part 4: (670+181) / 898 = 94.76%
Table.8. The testing accuracy for the our prototype via 6-fold cross-validation
Output / desired
Table.9. Confusion matrix for part 1 via 6-fold cross-validation method
Â Output / desiredÂ
Table.10. Confusion matrix for part 2 via 6-fold cross-validation method
Output / desiredÂ
Table.11. Confusion matrix for part 3 via 6-fold cross-validation method
Output / desired
Easy patient's tracheal
Hard Patient's tracheal
Easy patient's tracheal
Hard Patient's tracheal
Table.12. Confusion matrix for part 4 via 6-fold cross-validation method
From the confusion matrix of the first part shown in Table 9, 201 patients with ASA physical status 1 among 219 patients, 335 patients with ASA physical status 2 among 395, 207 patients with ASA physical status 3 among 232 patients and 43 patients with ASA physical status 4 among 52 patients were recognized correctly by the SVM classifier.
From the confusion matrix of the second part shown in Table 10 we remark that 723 patients accepted for surgery among 762 patients and 98 patients refused for surgery among 136 patients were recognized correctly.
From the confusion matrix for the third part shown in Table 11 we have 165 patients who general anesthesia technique is the best for surgery among 198 patients and 673 who local anesthesia technique is the best for surgery among 700 patients were recognized correctly by the SVM classifier.
From the confusion matrix for the fourth part shown in Table 12, 670 patients who tracheal intubation is easy among 700 patients and 181 who tracheal intubation is hard among 198 patients were recognized correctly by the classifier.
Our prototype gives a medical decision support system based on SVM for helping Doctors Specialized in Anesthesia in pre anesthetic consultation into four steps. The first one is the detection of ASA physical status, the second to choose if the patients are accepted or refused for surgery, the third is detection of the best anesthetic techniques (general or local), and finally to determine the patient's tracheal intubation.
A pre-anesthetic database consisting of 898 patients medical cases collected locally from different hospitals and privates clinical of western Algeria. The system has been developed with 30 input features and 10 classes.
Furthermore, the robustness of the proposed system was examined using 6-fold cross-validation method and results showed that the SVM-based decision support system could achieve average classification accuracy at 87.52% in the first part multiclasse,91.42% for the second part, 93.31% for the third part and finally 94.76% for the forth classifier.
The results obtained are promising and we wish to ameliorate our databases and to test other techniques of classification for given more precise output.