1. Introduction
Proposal techniques that reduce financial costs in the diagnosis and treatment of animal diseases are welcome. Among the poorest peoples in several parts of the world, there are one of the most severe forms of leishmaniasis, the visceral leishmaniasis (VL), also known as kalazar. VL is a life-threatening disease caused by Leishmania parasites, which are transmitted by female sandflies. VL causes fever, weight loss, spleen and liver enlargement, and, if not treated, death. People with both visceral leishmaniasis and HIV are difficult to cure [1]. The World Health Organization estimates that from 700,000 to 1,000,000 new VL cases occur annually [2]. VL is present in 88 countries, 22 of which are in the Americas. It is estimated that Brazil handles 90% of VL cases in Latin America [2]. Catão. R.C. [3] says that there is a stable interrelationship between the pathogen, vectors and people (infected and susceptible) with the geographic space. With leishmaniasis, mappings can help to understand the dynamics of transmission and the behavior of vectors. Injuries may be confined to one location or reach larger areas. Therefore, knowledge of spatial patterns in the occurrence of the disease becomes important for case surveillance [4].
The VL diagnoses are made by laboratory exams like the ELISA (enzyme-linked immunosorbent assay) test. These tests are sometimes very expensive for most residents in poor countries demonstrating the need for technologies that can reduce cost. In pursuing this aim, machine learning techniques appear as one of the most efficient methods in detecting VL in infected dogs.
Machine learning (ML) techniques employ the principle of induction by induction, getting results and extrapolations from a particular set of examples [5,6,7]. The ML system can be defined as a multi-component system, with an interface, learning algorithm, data, infrastructure and hardware. The learning algorithm is classified in two major categories: Supervised and unsupervised. In supervised learning, knowledge of the external environment is presented by sets of examples as desired input and output, in which the ML algorithm extracts the knowledge representation from these examples. The aim is that the generated representation can produce correct outputs for new inputs not presented [5,6,7]. With unsupervised learning, the model will not receive the desired output. The goal is for the machine to extract information from the input variables in order to separate them into different classes [8]. Unsupervised learning is the most widely used type of machine learning [9] and regression models (supervised) are the most used predictive model types, and among them, logistic regression analysis is used for dichotomous outputs. Logistic regression accounts for or predicts values of a single result variable with information from one or more explanatory variables and can classify an observation into one of two or more classes [10]. Logistic regression is one of the most used analytical tools in social and natural sciences [11].
Several studies have used machine learning to diagnose canine diseases. Larius, G. et al. [12] developed a method for the diagnosis of canine visceral leishmaniasis based on Fourier-Transform Infrared Spectroscopy (FTIR spectroscopy) and machine learning, in which canine blood sera from twenty uninfected dogs, twenty Leishmania infantum and eight dogs infected with Trypanosoma evansi were analyzed. They used principal component analysis with machine learning algorithms and archived over 85% in diagnosing true positives. Reagan, KL et al. [13] also applied machine-learning techniques to aid in the diagnosis of Canine Hypoadrenocorticism (CH) using screening diagnosis by complete blood count and serum chemistry panel. The database used was 908 control dogs with suspected CH and 133 dogs with confirmed CH. A driven tree algorithm was trained and tested to assess performance, with a sensitivity of 96.3%, and a specificity of 97.2%. A lymph node parasite load prediction model from clinical data in dogs with visceral leishmaniasis by artificial neural networks and machine learning was presented in [14]. In this study, 55 (fifty-five) dogs from seven regions of the states of Bahia, Minas Gerais, São Paulo and Distrito Federal, with 35 infected dogs and twenty control dogs, archived accuracy of 78% in the analyses performed. In the research carried out in [15], four machine learning algorithms were used to predict the diagnosis of Cushing’s syndrome, using structured clinical data from the VetCompass program in the UK. Cushing’s syndrome, which is an endocrine disease in dogs, negatively affects the quality of life of affected dogs. Machine learning methods could classify the recorded Cushing syndrome diagnoses, with a predictive result for regression with a sensitivity of 0.71 and a specificity of 0.82. We can notice that in these works all researchers used some kind of laboratory exam.
In this work, we propose the use of machine learning methods to make a machine that predicts if a certain animal has or does not have canine visceral leishmaniasis based only on physical examination of it. For that, four machine learning algorithms were used and the best one was chosen in the classification case of VL. Data were used from canine clinical exams realized in a certain region of the state of Maranhão, Brazil, to be used in the models.
This work is divided into three more sections besides this introduction. In Section 2, the materials and methods used in the work are discussed, in Section 3 the results are presented, and in Section 4 the final considerations are given.
2. Materials and Methods
According to Bassert, J.M. et al. [16] “History and physical examination are the first steps in the technician’s observation of any patient or group of patients. The information obtained from these processes serves as the basis for all subsequent assessments and interventions. It is essential that veterinary technicians can get complete and accurate historical information in the assessments of each patient and group. Similarly, good physical examination skills allow the quick identification of significant problems, followed by appropriate therapeutic measures”. The physical examination includes a professional assessment of the patient’s health and well-being.
In this work, the database was created from existing clinical examination records on 340 (three hundred and forty) dogs (cases: n = 177, non-cases: n = 163). We got seventeen variables that describe the dog’s characteristics, as seen in Table 1. These variables, according to the veterinaries, are variables that they observe in a first view of the animal suspected of having VL: Sex, presence of ectoparasites, nutrition, lymph nodes, mucosal color, bleeding, coat, muzzle and/or ear injury, nails, presence of skin lesion, depigmentation, alopecia, eye secretion, blepharitis, proximity to the forest and the ELISA (enzyme-linked immunosorbent assay) test results. With that information from veterinaries, our initial step was to use these variables to train the models. In Table 1 we also can see the p = value of a correlation test between the levels of the variables.
The ELISA test is a quantitative serological method making up a tool used both for analysis of clinical suspicion and for confirming the diagnosis of leishmaniasis. Confirmation occurs through the detection of immunoglobulinG (IgG) in the serum of suspected dogs. This exam is chosen because of its specificity and sensitivity [17,18]. The ELISA test’s performance is related not only to the type of antigen used but also to the clinical state manifested by the dog [19]. This test is considered the golden standard test in the diagnosis of leishmaniasis and it is the confirmatory test recommended by the Brazilian Ministry of Health [20]. In this work, the results of the ELISA (positive or not) test are used as the dependent (target) variable.
Data collection was performed out in certain regions of the west of the state of Maranhão (1°59′–4°00 S and 44°21′–45°33′ W), which is low in the Human Development Index (HDI) [21] (Figure 1).
2.1. Model Selection and Variable Selection
The target variable, ELISA test results, is dichotomous and Logistic Regression (LR) appears as a good choice for the learning algorithm [22,23,24,25]. Four algorithms were tested to choose the best model: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Naïve Bayes (NB) and LR, as well as the K-nearest neighbor classifier, which is based on the characteristic of the k-nearest neighbor of a new point (sample) to classify it. In this work, the best results were achieved with k = 10.
Naïve Bayes classifier is based on the assumption of independence between the variables of the problem. The NB model performs a probabilistic classification of an unclassified sample to put it in the most likely class.
Support vector machine is a high-performance model for nonlinear problems, not biased by outliers and not sensitive to them. It includes Support Vector Classification (SVC) and Support Vector Regression (SVR) [26].
For each model, we applied a recursive feature elimination with cross-validation as a preprocessing step [27,28] to select the best variables. For the SVM model, the excluded variables were age, lymph nodes and eye secretion. For LR, NB and KNN model the variables excluded were age, condition and depigmentation. The algorithms were trained and tested with the dataset containing only the best variables.
Regression models are one of the most important statistical tools in the statistical analysis of data for modelling relationships between variables. These models aim to detect the relationship between one or more explanatory variables, and response, or dependent variables. One of the particular cases of generalized logistic models is the one in which the response variable has only two categories of dichotomized values (0 or 1) [10].
Logistic regression aims to model, from a set of observations, the logistic relationship (probability distribution) between a dichotomous response variable and a series of numerical explanatory variables, which can be continuous, discrete and categorical [10,22]. The idea is to use the logistic expression given by:
y = (1 + e−z)−1(1)
where z = a0 + aTX, X is an m × n matrix containing m examples with n features, y is an m × 1 array of 0 and 1, and a is an m × 1 vector containing the parameters of the system, which will be inferred by the learning algorithm. This inference is done by an interactive task aiming to minimize the error between the actual values and the inferred values of y. After obtaining the parameter vector, a, one can infer a value for a new sample. The classification is made in the following way: The learning algorithm will, for each example, determine a number by equation (1), which represents the probability of y = 1, and if this number is equal or greater than 0.5 will put y = 1, or 0 otherwise. With the parameter vector a, we can assign a number for each new dog feature vector shown to the system.For example, we can assign the 240 × 14 feature matrix (after variables selection), X, containing the values of the 14 features for each one of the 240 dogs and a 240 × 1 vector, y, containing the values 0 or 1 if the dog does not have or has the disease. We separate this sample into two parts: 80% for training and 20% for test. For the training set, we present the learning algorithm with a 192 × 14 matrix and a corresponding 192 × 1 vector y. In this learning phase, the algorithm will estimate the parameters ai, i = 0, …, 14. With the estimated vector a, we can get z and put it in Equation (1) to estimate the values of the y for each sample in the test set. With the inferred and actual values, we can get the confusion matrix to get the metrics explained in the next section. In this work, after the training phase, we got a = [0.00401544, −0.3851127, 0.25599153, −0.05313171, 0.54447301, 0.36488774, −0.21396936, 0.15184775, 0.28558144, −0.11643821, 0.47422156, 0.398408060, −0.34161375, −1.24370334] T and a0 = 0.30852909.
2.2. Diagnostic Test
Diagnosing a disease is a delicate matter because the lives of patients are at stake, whether they are humans, dogs or even plants. The tools used in the diagnostic process are tests based on measurements made on patients, whether quantitative or qualitative, called clinical tests or diagnostic tests [23,24,25].
These tools have become so important and widespread that there are large industries and laboratories entirely dedicated to the production of increasingly accurate, rapid and inexpensive diagnostic tests. Tests can be misleading, especially where there may be a problem with a biological system. Before a test is used as an aid in the diagnosis of a certain disease, its potential for error must be evaluated [23,24,25].
Technological proposals to reduce the financial costs of treating diseases and the use of general laboratory tests are of great interest. These technologies act like screening tests leaving laboratory tests to be performed only on beings with a high probability of disease presence. The machine learning techniques can help identify sick individuals with a reasonable statistical probability of true positives.
Yang Xin et al. [26] say: “The evaluation model is a very important part of the machine-learning mission”. In this work, we follow their steps to evaluate our proposal, using the metrics obtained from the confusion matrix. The confusion matrix is shown in Table 2.
Further, the following metrics can be calculated from the confusion matrix [26]:
Accuracy: (TN + TP)/(TN + FP + FN + TP). This measures the fraction of correct predictions.
Sensitivity or Recall: (TP)/(TP + FN). This measures the ability of the test to correctly identify individuals who have the disease. It measures the probability of the test getting a positive result given that the true condition is present. This is the most important metric in screening because a negative result in a test with high sensitivity is useful for excluding the existence of the condition.
Specificity: TN/(TN + FP). This is the ability of the test in correctly identify individuals who do not have the disease;
The Positive Predictive Value or Precision: TP/(TP + FP). This measures the probability of the dog having the disease knowing that the test result is positive.
The Negative Predictive Value: TN/(TN + FN). This measures the probability of the dog has not had the disease knowing that the test result is negative.
The Positive Likelihood Ratio (LR+): Sensitivity/(1 − Specificity). This shows that for a value greater than 1 (one), the positive test is more likely to occur in dogs with the disease than in those without the disease;
The Negative Likelihood Ratio (LR−): (1 − Sensitivity)/Specificity. This shows that for a value greater than 1 (one), the negative test is more likely to occur in dogs with the disease than in those without the disease.
The Area Under Curve (AUC): from the Receiver Operating Characteristic (ROC) curve: This is performed to identify how good the model developed is at distinguishing between two parameters, the true positive rate and the false negative rate. Models with 100% correct predictions have an AUC of 1.
2.3. Canine Visceral Leishmaniasis
Leishmaniasis belongs to the group of diseases caused by a parasitic protozoan of the genus Leishmania which is transmitted to humans and other various mammals through the bite of females of a hematophagous insect dipterans of the Psychodidae family, subfamily Phlebotominae, known generically as sandflies, playing the role of a vector in the disease cycle [2,29,30]. The World Health Organization has included leishmaniasis as one of the six most important diseases in the world. Even included in this list, leishmaniasis is considered a neglected disease. It is related to the poverty of people with deteriorating housing and bad sanitation conditions and is common in regions with a low economic development index.
Furtado, A. S. et al. [31] say that Maranhão showed an expansion of cases of human leishmaniasis in the period from 2000 to 2009. From 1999 to 2005, the state led the number of confirmed cases of the disease in Brazil. In the year 2019, according to data from the Notification System of the Health Surveillance Secretariat of the Ministry of Health, 430 confirmed cases of visceral leishmaniasis in humans were reported [1], which shows the importance of research in early detection of main vectors of disease spread.
3. Results and Discussions
The database was split into eighty per cent for training and twenty per cent for testing. The training set was used to determine the parameters of the learning algorithm and the test set was used to validate the models by the metrics described above. Table 3 shows the results for the models used.
From Table 3, one can see that the LR model got better results than the others did, for instance, accuracy of 75%, sensitivity of 84% and negative predictive value of 83%. These are good results because they assure us great security in that the dog predicted as not having the disease does not actually have it; then it is not necessary to carry out a laboratory exam on it. A positive predictive value of 0.69 means that we have approximately 70% certainty that the dogs tested as positive really have the disease. The LR+ equal to 2.53 means that a dog with the disease is 2.53 times more likely to have a positive test than one without the disease. The LR− equal to 0.23 means that a dog without the disease is, approximately, four (0.25) times more likely to test negative than those with the disease. Thus, the logistic regression model shows a good ability in rejecting false negatives.
The AUC of 0.77 (Figure 2) shows the test’s discriminatory ability to distinguish between dogs with and without the disease.
From those results, one can observe that the logistic regression model can act as an efficient screening method for dogs with canine visceral leishmaniasis based only on their visualization and thus reducing the cost in laboratory exams.
As an attempt to understand the type of correctly and not correctly classified samples, for the LR model, we get the descriptive characteristics of each variable for the four classifications: TN, FP, FN and TP.
In Table 4 we have the confusion matrix. One can see that this model got five false negative and 12 false positive samples.
Table 5 shows the samples classified as false negative. One can see that 100% of the samples present the following characteristics: Normal mucosal color, no bleeding, augmented nails, no presence of skin lesion, no eye secretion and no blepharitis.
Table 6 shows the samples classified as false positive. One can see that 100% of the samples present the following characteristics: No bleeding and no eye secretion.
Table 7 shows the samples classified as true negative. One can see that samples possess the following characteristics: No presence of skin lesion and no eye secretion.
Table 8 shows the samples classified as true positive. One can see that a majority of true positive samples have no presence of ectoparasites, enlarged lymph nodes, no presence of bleeding, no presence of skin lesion, no eye secretion, no blepharitis and no proximity to the forest.
4. Final Considerations
In this work, four machine learning models were tested as an initial method in veterinary care to identify dogs with canine visceral leishmaniasis based only on visual inspection of the animal. For that, we got clinical dates from 340 dogs with eighteen variables. These variables were chosen based on veterinary professionals’ experiences and for each model the best variables were selected to predict the results. The models tested were logistic regression, support vector machine, K-nearest neighbor, and Naïve Bayes. The logistic regression model, using fourteen variables after the variable selection procedure, got the best metrics: Accuracy of 75%, sensitivity of 84%, specificity of 67%, positive likelihood ratio of 2.53 and negative likelihood ratio of 0.23. This model enables cost reduction in this type of care and can become a useful tool to screen this disease, contributing to the improvement of urban public health.
Conceptualization, L.S.B., A.L.A.S. and S.A.M.; methodology, E.E.C.S., P.F.S.J. and A.F.L.J.J.; software, A.F.L.J.J., T.S.F. and G.O.L.; validation, L.S.B., A.L.A.S. and S.A.M.; formal analysis, G.O.L., P.F.S.J., E.E.C.S.; investigation, L.S.O.C.; resources, V.S.A.; data curation, E.E.C.S., P.F.S.J.; writing—original draft preparation, T.S.F.; writing—review and editing, P.F.S.J.; visualization, P.F.S.J.; supervision, E.E.C.S.; project administration, R.C.S.F.; funding acquisition, C.A.M.C. All authors have read and agreed to the published version of the manuscript.
This work was supported by Fundação de Pesquisa do Estado do Amazonas–FAPEAM under POSGRAD program EDITAL N 008/2021.
The animal protocols used in this work were evaluated and approved by the Animal Ethics and Experimentation Committee of the Center for Agricultural Sciences of the State University of Maranhão, under Process nº 01200.002200/2015-06, and are in accordance with Law 11.794/2008 of the Republic Federation of Brazil.
The data used in the study were collected and authorized for use by the ethics committee in animal experimentation of the veterinary medicine course at the Agricultural Sciences Center of the State University of Maranhão, with process no. 037/2017, opinion No. 037/2017.
The data used in this work are not available for consultation on the website, being the property of the Graduating Program in Animal Sciences of the State University of Maranhão.
This work was supported by Fundação de Pesquisa do Estado do Amazonas–FAPEAM under POSGRAD program EDITAL N 008/2021. We greatly appreciate the CAPES, CNPq, FAPEMA, UEMA, FAPEAM, UFAM, FAPESQ-PB, UFCG, and UFMA, the Department of Zoonosis Control of the State Department of Health-SES/MA and the Central Public Health Laboratory through the Nucleus of Endemics, serology sector (IOC/LACEN-MA) by supporting and funding this project, without which this work would not be possible.
The authors declare no conflict of interest.
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Descriptive statistics and univariable associations of features included in machine learning prediction of the canine visceral leishmaniasis (cases: n = 177; non-cases: n = 163).
| Variable | Category | Non-Cases | Cases | p-Value |
|---|---|---|---|---|
| Sex | Female | 74 | 74 | 0.505 |
| Male | 89 | 103 | ||
| Age (months) | Mean/standard deviation | 34.39/30.8 | 44.03/36.45 | 0.009 |
| Condition | Apathetic | 19 | 27 | 0.346 |
| Active | 144 | 150 | ||
| Presence of ectoparasites | No | 127 | 151 | 0.078 |
| Yes | 36 | 26 | ||
| Nutrition | Normal | 119 | 113 | 0.058 |
| Thin | 41 | 53 | ||
| Skinny | 3 | 11 | ||
| Lymph nodes | Normal | 25 | 27 | 0.983 |
| Enlarged | 138 | 150 | ||
| Mucosal color | Normal | 121 | 123 | 0.332 |
| Pale | 42 | 54 | ||
| Bleeding | No | 156 | 162 | 0.118 |
| Yes | 7 | 15 | ||
| Coat | Normal | 87 | 65 | 0.007 |
| Regular | 44 | 70 | ||
| Bad | 32 | 42 | ||
| Muzzle and/or ear injury | No | 133 | 118 | 0.002 |
| Yes | 30 | 177 | ||
| Nails | Augmented | 127 | 100 | <0.001 |
| onychogryphosis | 36 | 77 | ||
| Presence of skin lesion | No | 153 | 161 | 0.314 |
| Yes | 10 | 16 | ||
| Depigmentation | No | 162 | 177 | 0.479 |
| Yes | 1 | 0 | ||
| Alopecia | No | 116 | 89 | <0.001 |
| Yes | 47 | 88 | ||
| Eye secretion | No | 159 | 166 | 0.115 |
| Yes | 4 | 11 | ||
| Blepharitis | No | 145 | 157 | 0.94 |
| Yes | 18 | 20 | ||
| Proximity to the forest | No | 77 | 141 | <0.001 |
| Yes | 86 | 36 |
Indications of the confusion matrix.
| Predicted as Negative | Predicted as Positive | |
|---|---|---|
| Labeled as Negative | True Negative (TN) | False Positive (FP) |
| Labeled as Positive | False Negative (FN) | True Positive (TP) |
Test Metrics of the models tested. One can see that LR got the best metrics.
| NB | KNN | SVM | LR | |
|---|---|---|---|---|
| Accuracy | 0.63 | 0.63 | 0.69 | 0.75 |
| Sensitivity (Recall) | 0.56 | 0.56 | 0.84 | 0.84 |
| Specificity | 0.69 | 0.69 | 0.56 | 0.67 |
| Positive Predictive Value | 0.62 | 0.62 | 0.63 | 0.69 |
| Negative Predictive Value | 0.64 | 0.64 | 0.80 | 0.83 |
| LR+ | 1.84 | 1.84 | 1.90 | 2.53 |
| LR− | 0.63 | 0.63 | 0.28 | 0.23 |
| AUROC | 0.71 | 0.71 | 0.70 | 0.77 |
Confusion matrix.
| Predicted as Negative | Predicted as Positive | |
|---|---|---|
| Labeled as Negative | 24 | 12 |
| Labeled as Positive | 5 | 27 |
Frequency of the variables for the false negatives samples.
| Variable | Category | Frequency (%) |
|---|---|---|
| Sex | Female | 20 |
| Male | 80 | |
| Presence of ectoparasites | No | 80 |
| Yes | 20 | |
| Nutrition | Normal | 60 |
| Thin | 40 | |
| Skinny | 0 | |
| Lymph nodes | Normal | 20 |
| Enlarged | 80 | |
| Mucosal color | Normal | 100 |
| Pale | 0 | |
| Bleeding | No | 100 |
| Yes | 0 | |
| Coat | Normal | 20 |
| Regular | 60 | |
| Bad | 20 | |
| Muzzle and/or ear injury | No | 80 |
| Yes | 20 | |
| Nails | Augmented | 100 |
| onychogryphosis | 0 | |
| Presence of skin lesion | No | 100 |
| Yes | 0 | |
| Alopecia | No | 80 |
| Yes | 20 | |
| Eye secretion | No | 100 |
| Yes | 0 | |
| Blepharitis | No | 100 |
| Yes | 0 | |
| Proximity to the forest | No | 20 |
| Yes | 80 |
Frequency of the variables for the false positive samples.
| Variable | Category | Frequency (%) |
|---|---|---|
| Sex | Female | 58 |
| Male | 42 | |
| Presence of ectoparasites | No | 0.92 |
| Yes | 0.08 | |
| Nutrition | Normal | 83 |
| Thin | 17 | |
| Skinny | 0 | |
| Lymph nodes | Normal | 8 |
| Enlarged | 92 | |
| Mucosal color | Normal | 75 |
| Pale | 25 | |
| Bleeding | No | 100 |
| Yes | 0 | |
| Coat | Normal | 50 |
| Regular | 30 | |
| Bad | 20 | |
| Muzzle and/or ear injury | No | 83 |
| Yes | 17 | |
| Nails | Augmented | 58 |
| onychogryphosis | 42 | |
| Presence of skin lesion | No | 83 |
| Yes | 17 | |
| Alopecia | No | 42 |
| Yes | 58 | |
| Eye secretion | No | 100 |
| Yes | 0 | |
| Blepharitis | No | 92 |
| Yes | 8 | |
| Proximity to the forest | No | 92 |
| Yes | 8 |
Frequency of the variables for the true negatives samples.
| Variable | Category | Frequency (%) |
|---|---|---|
| Sex | Female | 42 |
| Male | 58 | |
| Presence of ectoparasites | No | 75 |
| Yes | 25 | |
| Nutrition | Normal | 83 |
| Thin | 17 | |
| Skinny | 0 | |
| Lymph nodes | Normal | 8 |
| Enlarged | 92 | |
| Mucosal color | Normal | 67 |
| Pale | 33 | |
| Bleeding | No | 92 |
| Yes | 8 | |
| Coat | Normal | 62 |
| Regular | 21 | |
| Bad | 17 | |
| Muzzle and/or ear injury | No | 96 |
| Yes | 4 | |
| Nails | Augmented | 96 |
| onychogryphosis | 4 | |
| Presence of skin lesion | No | 0 |
| Yes | 100 | |
| Alopecia | No | 92 |
| Yes | 8 | |
| Eye secretion | No | 100 |
| Yes | 0 | |
| Blepharitis | No | 96 |
| Yes | 4 | |
| Proximity to the forest | No | 21 |
| Yes | 79 |
Frequency of the variables for the true positive samples.
| Variable | Category | Frequency (%) |
|---|---|---|
| Sex | Female | 30 |
| Male | 70 | |
| Presence of ectoparasites | No | 93 |
| Yes | 7 | |
| Nutrition | Normal | 67 |
| Thin | 22 | |
| Skinny | 11 | |
| Lymph nodes | Normal | 12 |
| Enlarged | 88 | |
| Mucosal color | Normal | 63 |
| Pale | 37 | |
| Bleeding | No | 89 |
| Yes | 11 | |
| Coat | Normal | 26 |
| Regular | 37 | |
| Bad | 37 | |
| Muzzle and/or ear injury | No | 63 |
| Yes | 37 | |
| Nails | Augmented | 44 |
| onychogryphosis | 56 | |
| Presence of skin lesion | No | 85 |
| Yes | 15 | |
| Alopecia | No | 40 |
| Yes | 60 | |
| Eye secretion | No | 85 |
| Yes | 15 | |
| Blepharitis | No | 93 |
| Yes | 7 | |
| Proximity to the forest | No | 97 |
| Yes | 3 |
References
1. Cadernos de Saúde Pública. DATASUS. 2020; Available online: http://tabnet.datasus.gov.br/cgi/tabcgi.exe?sinannet/cnv/leishvma.def (accessed on 29 September 2021).
2. World Health Organization. Leishimaniasis. Available online: https://www.who.int/data/gho/data/themes/topics/topic-details/GHO/leishmaniasis (accessed on 5 January 2022).
3. Catão, R.C. Dengue No Brasil: Abordagem Geográfica na Escala Nacional; Cultura Acadêmica: São Paulo, Brazil, 2012.
4. Siquera, S.C.F. Análise Espacial da Dengue no Estado de Mato Grosso no Período de 2007 A 2009. Master’s Thesis; Universidade Federal de Mato Grosso: Cuiabá, Brazil, 2011.
5. Bishop, C.M. Pattern Recognition and Machine Learning; Springer: Berlin/Heidelberg, Germany, 2006.
6. Mitchell, T. Machine Learning; McGraw-Hill Science: New York, NY, USA, 1997.
7. Haykin, S. Neural Networks and Learning Machines; 3rd ed. Pearson: San Antonio, TX, USA, 2009.
8. Algore, M. Machine Learning with Python: The Definitive Tool to Improve Your Python Programming and Deep Learning to Take You to the Next Level of Coding and Algorithms Optimization; Kindle: Zürich, Switzerland, 2021.
9. Alpaydin, E. Introduction to Machine Learning; MIT Press: Cambridge, MA, USA, 2020.
10. Hoffmann, J.P. Linear Regression Models: Applications in R; Chapman and Hall/CRC: Boca Raton, FL, USA, 2021.
11. Speelman, D. Logistic regression: A confirmatory technique for comparisons in corpus linguistics. Corpus Methods Semant. Quant. Stud. Polysemy Synon.; 2014; 43, pp. 487-533. [DOI: https://dx.doi.org/10.1075/hcp.43.18spe]
12. Larios, G.; Ribeiro, M.; Arruda, C.; Oliveira, S.; Canassa, T.; Baker, M.J.; Marangoni, B.; Ramos, C.; Cena, C. A new strategy for canine visceral leishmaniasis diagnosis based on FTIR spectroscopy and machine learning. J. Biophotonics; 2021; 14, e202100141. [DOI: https://dx.doi.org/10.1002/jbio.202100141] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34423902]
13. Reagan, K.L.; Reagan, B.A.; Gilor, C. Machine learning algorithm as a diagnostic tool for hypoadrenocorticism in dogs. Domest. Anim. Endocrinol.; 2020; 72, 106396. [DOI: https://dx.doi.org/10.1016/j.domaniend.2019.106396] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32006871]
14. Torrecilha, R.B.P.; Utsumoniya, Y.T.; Batista, L.F.S.; Bosco, A.M.; Nunes, C.M.; Ciarlini, P.C. Prediction of lymph node parasite load from clinical data in dogs with leishmaniasis: An application of radial basis artificial neural networks. Vet. Parasitol.; 2017; 234, pp. 13-18. [DOI: https://dx.doi.org/10.1016/j.vetpar.2016.12.016] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28115177]
15. Schofield, I.; Brodbelt, D.C.; Kennedy, N.; Niessen, S.J.M.; Church, D.B.; Geddes, R.F.; O’Neill, D.G. Machine-learning based prediction of Cushing’s syndrome in dogs attending UK primary-care veterinary practice. Sci. Rep.; 2021; 11, 9035. [DOI: https://dx.doi.org/10.1038/s41598-021-88440-z] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33907241]
16. Bassert, J.M.; Beal, A.D.; Samples, O.M. McCurnin’s Clinical Textbook for Veterinary Technicians; Elsevier: Amsterdam, The Netherlands, 2018.
17. Alves, W.A. Leishmaniose visceral americana: Situação atual no Brasil Leishmaniasis: Current situation in Brazil. World Health; 2009; 6, pp. 25-29.
18. Fonseca, T.H.S.; Faria, A.R.; Leite, H.M.; Da Silveira, J.A.G.; Carneiro, C.M.; Andrade, H.M. Chemiluminescent ELISA with Multi-Epitope Proteins to Improve the Diagnosis of Canine Visceral Leishmaniasis. Vet. J.; 2019; 253, 105387. [DOI: https://dx.doi.org/10.1016/j.tvjl.2019.105387] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31685139]
19. Faria, A.R.; De Andrade, H.M. Diagnóstico da Leishmaniose Visceral Canina: Grandes avanços tecnológicos e baixa aplicação prática. Rev. Pan-Amaz. Saúde; 2012; 3, 11. [DOI: https://dx.doi.org/10.5123/S2176-62232012000200007]
20. Verotti, M.P. Clarifications on Replacement of the Diagnostic Protocol for Canine Visceral Leishmaniasis. Technical Note n. 1, General Coordination of Communicable Diseases/General Coordination of Public Health Laboratories; Department of Communicable Disease Surveillances, Department of Health Surveillance, Ministry of Health: Brasilia, Brazil, Available online: http://www.sgc.goias.gov.br/upload/arquivos/2012-05/nota-tecnica-no.-1-2011_cglab_cgdt1_lvc.pdf (accessed on 18 March 2021).
21. IBGE. Synopsis of the 2010 Population Cesus do Censo Demográfico 2010. Rio de Janeiro. Available online: http://www.ibge.gov.br (accessed on 18 March 2021).
22. Kleinbaum, D.G. Logistic Regression; Springer: New York, NY, USA, 2002.
23. Vaden, S.L.; Knoll, J.S.; Smith, F.W.K., Jr.; Tilley, L.P. Blackwell’s Five-Minute Veterinary Consult: Laboratory Tests and Diagnostic Procedures: Canine and Feline; 5th ed. Wiley: Hoboken, NJ, USA, 2009.
24. Hendrix, C.M. Diagnostic Parasitology for Veterinary Technicians; 4th ed. Elsevier-Mosby: Amsterdam, The Netherlands, 2012.
25. Neuber, A.; Nuttall, T. Diagnostic Techniques in Veterinary Dermatology: A Manual of Diagnostic Techniques; Wiley Blackwell: Hoboken, NJ, USA, 2017.
26. Xin, Y.; Kong, L.; Liu, Z.; Chen, Y.; Li, Y.; Zhu, H.; Gao, M.; Hou, H.; Wang, C. Machine Learning and Deep Learning Methods for Cybersecurity. IEEE Access; 2018; 6, pp. 35365-35381. [DOI: https://dx.doi.org/10.1109/ACCESS.2018.2836950]
27. Brownlee, J. Data Preparation for Machine Learning: Data Cleaning, Feature Selection, and Data Transform in Python; Machine Learning Mastery: New York, NY, USA, 2020.
28. Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V. et al. Scikit-Learn: Machine Learning in Python. J. Mach. Learn. Res.; 2011; 12, pp. 2825-2830.
29. Neves, D.P. Parasitologia Humana; 13th ed. Editora Atheneu: São Paulo, Brazil, 2016.
30. Drugs for Neglected Diseases Institute. Viceral Leishmaniasis: Symptoms, Transmission, and Treatments for Visceral Leishmaniasis. 2020; Available online: https://dndi.org/diseases/visceral-leishmaniasis/facts/ (accessed on 14 January 2022).
31. Furtado, A.S.; Nunes, F.B.; Santos, A.M.; Caldas, A.J. Space-time analysis of visceral leishmaniasis in the State of Maranhão, Brazil. Ciências E Saúde Coletiva; 2015; 20, pp. 35-42. [DOI: https://dx.doi.org/10.1590/1413-812320152012.01672015]
© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
; Silva Junior, Paulo F 1
; Bastos, Luciana S 2 ; Silva, Ana L A 2 ; Melo, Solange A 3 ; Cruz, Carlos A M 4
; Aquino, Vivianne S 4 ; Castro, Luís S O 4 ; Lima, Guilherme O 5 ; Freire, Raimundo C S 6
1 Graduating Program in Computation Engineering and Systems, State University of Maranhão, São Luís 65690-000, Brazil;
2 Graduating Program in Animal Sciences, State University of Maranhão, São Luís 65690-000, Brazil;
3 Graduating Program in Animal Health Defense, State University of Maranhão, São Luís 65690-000, Brazil;
4 Graduation Program in Electrical Engineering, Federal University of Amazonas, Manaus 69067-005, Brazil;
5 Graduation Program in Electrical Engineering, Federal University of Maranhão, São Luís 65690-000, Brazil;
6 Graduation Program in Electrical Engineering, Federal University of Campina Grande, Campina Grande 58428-830, Brazil;