A novel convolutional neural network–Fuzzy-based diagnosis in the classification of dental pulpitis

Rahulsinh Bhupendrasinh Chauhan; Tejas V Shah; Deepali H Shah; Tulsi Jaduvirsinh Gohil

doi:10.4103/aihb.aihb_50_22

ORIGINAL ARTICLE

Year : 2023 | Volume : 13 | Issue : 1 | Page : 79-86

A novel convolutional neural network–Fuzzy-based diagnosis in the classification of dental pulpitis

Rahulsinh Bhupendrasinh Chauhan¹, Tejas V Shah², Deepali H Shah³, Tulsi Jaduvirsinh Gohil⁴
¹ Department of Bio-Medical, Government Polytechnic Gandhinagar, Gujarat Technological University, Ahmedabad, Gujarat, India
² Department of Instrumentation and Control, L D College of Engineering, Gujarat Technological University, Ahmedabad, Gujarat, India
³ Department of Instrumentation and Control, Government Engineering College Gandhinagar, Gujarat Technological University, Ahmedabad, Gujarat, India
⁴ Nilkanth Dental Care, Pethapur, Gandhinagar, Gujarat, India

Date of Submission	08-Mar-2022
Date of Acceptance	29-Apr-2022
Date of Web Publication	17-Oct-2022

Correspondence Address:
Prof. Rahulsinh Bhupendrasinh Chauhan
Department of Bio-Medical, Government Polytechnic Gandhinagar, Gandhinagar, Gujarat
India

Source of Support: None, Conflict of Interest: None

Check

DOI: 10.4103/aihb.aihb_50_22

Abstract

Introduction: This study presents a computer-aided decision-making system based on the convolutional neural network (CNN)–fuzzy approach. According to the literature, there is a lack of coherence amongst dentists in diagnosing reversible or irreversible pulpitis. As a result, the goal of this research is to assist dentists in accurately diagnosing pulpitis. Materials and Methods: A rigorous algorithm that relies on CNN-fuzzy logic has been designed to handle inaccurate and ambiguous values of dental radiographs, as well as signs and symptoms of pulpitis. To begin, the probability of cavity for each class was determined using an independently designed CNN approach, which was then applied in combination with symptoms associated with pulpitis to a fuzzy knowledge base with 665 rules and the Mamdani inference algorithm to diagnose pulpitis and make recommendations to the dentist. Results: The CNN-fuzzy approach's results are compared to the dentists' recommendations. With the assistance of five professional dentists, the sensitivity, specificity, precision, accuracy, f1 score and Matthews correlation coefficient are calculated from 100 randomly generated sample cases. The CNN-fuzzy approach has a 94% accuracy, which is 7% higher than expert prediction. It is observed that the proposed approach produces results that are consistent with the dentists' diagnoses. Conclusion: The accuracy of the proposed computer-aided decision-making system for pulpitis increases dentists' confidence in diagnosing reversible and irreversible pulpitis and reduces false diagnoses due to ambiguous values of dental radiographs, signs and symptoms.

Keywords: Convolutional neural networks, dental diagnosis, fuzzy logic, irreversible pulpitis, reversible pulpitis

How to cite this article:
Chauhan RB, Shah TV, Shah DH, Gohil TJ. A novel convolutional neural network–Fuzzy-based diagnosis in the classification of dental pulpitis. Adv Hum Biol 2023;13:79-86

How to cite this URL:
Chauhan RB, Shah TV, Shah DH, Gohil TJ. A novel convolutional neural network–Fuzzy-based diagnosis in the classification of dental pulpitis. Adv Hum Biol [serial online] 2023 [cited 2023 Mar 27];13:79-86. Available from: https://www.aihbonline.com/text.asp?2023/13/1/79/362792

Introduction

The most common non-contagious disease is an oral disease, which affects people in a variety of ways throughout their lives. Pulpitis is an inflammatory disorder of the dental pulp or the soft tissue in the centre of the tooth. According to a systematic review conducted by Rechenberg et al.^[1] in 2016, the most common cause of pulpitis is when bacteria irritate the dental pulp through an area of tooth decay, such as dental caries, trauma or injury to a tooth. These irritants typically cause reversible pulpitis first. If the pulp remains inflamed, the pulpitis becomes irreversible, and the pulp may eventually die. A dentist can diagnose pulpitis in a conventional inspection based on a person's symptoms, an analysis of the teeth and probably a radiographic examination. Other tests, such as a sensitivity test, tooth tap test and electric pulp test, may be performed in some cases by the dentist.

Radiographs are the most widely used diagnostic tool, and as new techniques have advanced, manual radiograph analysis has been overtaken by computer-aided diagnosis, ranging from traditional approaches to machine learning or deep learning-based approaches for various kinds of radiographs for diagnosis.

Several studies have been carried out to demonstrate image processing and machine learning or deep learning-based approaches for predicting dental caries in different types of radiographs. The most commonly used digital diagnostic tools for dental caries are panoramic radiographs,^{[2],[3],[4],[5],[6],[7]} periapical radiographs,^{[8],[9],[10],[11],[12]} photographs,^{[13],[14],[15]} bitewing radiographs^[16],[17] and near-infrared light transillumination images,^[18],[19] but the majority of studies have chosen periapical radiographs and panoramic radiographs for dental caries detection. Even so, neither of the studies focus on determining whether dental pulpitis is reversible or irreversible.

The objective of this article is to diagnose dental pulpitis as reversible or irreversible pulpitis using a novel convolutional neural network (CNN)–fuzzy-based approach that is more accurate than a manual approach to eliminate or reduce inappropriate diagnosis and to provide a second opinion. To the best of our knowledge, we believe we are the first group to work on a novel CNN-fuzzy-based approach for computer-aided diagnosis of reversible or irreversible pulpitis.

Our research question is how to accurately diagnose reversible or irreversible pulpitis in radiographs because they have extremely small length differences, making depth-wise dental pulpitis classification difficult. To respond, we used CNN techniques for dental caries classification and fuzzy-based approaches for combining symptoms with CNN results for pulpitis diagnosis.

Materials and methods

This section is divided into three sections: the first focuses on datasets and pre-processing, the second on CNN and the last on the fuzzy approach.

Dataset and pre-processing

The dataset used for training and validation of our CNN-based dental X-ray cavity prediction model consists of 228 of 428 periapical X-ray images, 114 of which had deep cavities and the other 114 of which had shallow cavities and were evaluated and labelled by three dental surgeons. Leftover X-ray images were used to test the proposed method. The CNN model used in this study was created in Python using the Keras library, and the fuzzy approach was implemented in MATLAB. To compensate for the limitations of the small training dataset, we used data augmentation to improve our CNN model and reduce bias and generalisation errors. This method increased the number of training datasets by generating new samples for training images without changing the image's attributes. To understand the data augmentation, a small volume of training images was visualised, and the dataset was augmented with vertical and horizontal shifts and rotations. Images were resampled to 224 × 224-pixel resolutions and normalised for mean to aid model convergence.

Convolutional neural network

CNNs have recently emerged as prominent machine learning algorithms for image-based dental diagnosis^{[20],[21],[22]} since they preserve complex features while inspecting input images. The system architecture is represented in [Figure 1]. The convolution layer is made up of a set of filters (kernel). It is an operation in which we move a small matrix of numbers (known as the kernel or filter) over our image, transform it based on the kernel values and then extract information by creating a new layer. Each layer represents one or more of the significant features or characteristics of the input image. Subsequent feature map values are calculated using the below formula, where f represents the input image and k represents our kernel.^[23] The indexes of the rows and columns of the result matrix are denoted by m and n, respectively.

Figure 1: Proposed CNN architecture for dental cavity classification. CNN: Convolutional neural network.

Click here to view

This CNN architecture has ten convolution layers and accepts a 244 × 244 dental image tensor as input. The first convolution layer then employs 5 × 5 kernel filters with stride 1 × 1 for a total of 32 such filters. The first layer's output is then normalised by subtracting the batch mean and dividing the batch standard deviation. The second layer, which accepts the first layer's output, employs 64 filters, and its output is normalised before being passed to a max-pooling layer with a stride of 2 × 2, reducing the input to half its original size 112 × 112. For all layers, the output of the pooling layer is routed through the ReLU activation feature. The gained non-linear output is now fed into the two identical type convolution layers with 3 × 3 × 64, 128 filters, and the stride value is the same 1 × 1. The obtained output is passed through a max-pooling layer with the same 2 × 2 strides, reducing the input to half its original size 56 × 56. Following ReLU activation, the output is routed to the fifth to seventh convolution layers, each with 256 filters and a kernel size of 3 × 3 × 128, with a 1 × 1 stride. The output is passed through a max-pooling layer, resulting in a tensor of shape 28 × 28. The output is applied to ReLU activation again before being fed into the eighth to tenth convolution layer, which has 512 filters, a kernel size of 3 × 3 × 256 and the same stride of 1 × 1. The tenth convolution's output is max-pooled, yielding a tensor with the shape 14 × 14 × 512 and a flattened tensor with 100,352 neurons. The weighed values that evolve as neurons demonstrate the similarity to the symptoms of deep and shallow cavities. The dropout layer is used here to control network overfitting by dropping values; during training, we used a dropout rate of 0.2. The fully connected layer transforms the 100,352 neuron tensor into the number of categories (deep and shallow cavities) to which the dental image belongs. This method produced an accuracy of 85%, which is approximately 11% higher than the traditional VGG16, VGG19 and Xception approaches.

Proposed fuzzy-based system

In this research, we propose a new approach for dental decision-making based on a CNN-fuzzy system, with the aim of assisting the dentist in diagnosing reversible or irreversible pulpitis. The system receives observable signs and symptoms of pulpitis as noted by the dentist as well as verbal descriptions provided by the patient, and the proposed CNN network's predicted probability of deep and shallow cavities, and the diagnosis result is immediately provided to the dentist for further consideration.

Approaches based on fuzzy logic have been established and widely used in medical fields.^{[24],[25],[26]} The term 'fuzzy logic' first appeared in the development of professor Lotfi A. Zadeh's theory of fuzzy sets in 1965 and 1988. Fuzzy logic is a type of multi-valued logic derived from fuzzy set theory that is used to deal with approximate instead of precise reasoning. It is a type of artificial intelligence that employs a set of membership functions and fuzzy rules as opposed to traditional bi-valued logic. Fuzzifier, Inference Engine, Rule Applier and Defuzzifier are the standard four main functional components of the fuzzy logic-based expert system. The inference engine makes use of a knowledge base made up of if-then rules.

Fuzzification and the development of fuzzy membership functions for input–output variables

Fuzzification is the process of defining membership functions for linguistic expressions. [Table 1] shows the membership functions for the linguistic expressions used in this study as input and output variables.

Table 1: Linguistic expressions for input–output variables

Click here to view

Membership functions are typically derived from datasets or probability density functions. However, in our case, the lack of an available dataset in dentistry makes the use of automated algorithms challenging. As a result, we gained the support of our experts to create the membership functions and specify the universe of discourses for all variables. The universe of discourses was decided to be in the range [0,1]. A triangular membership function was initially considered for all variables, but we were unable to reach an agreement with the experts, so we relied on their responses and opted for Gaussian and other non-linear membership functions such as S-mf and Z-mf as they are more appropriate for this model. The following is a brief description of the various input and output variables and their corresponding membership functions used in this study.

Deep and shallow cavity/caries

It is essential to understand the nature of the tooth caries, i.e., whether the cavity is deep or shallow, by determining the level of Cavity from X-rays using the proposed CNN-based approach. Let z denotes the deep or shallow cavity. As a result, the membership function for deep or shallow cavities for linguistic expressions (low probability, medium probability and high probability) will be considered, as shown in Eqs.(2-4).

Continuous pain, pus discharge and swelling around tooth/gums

The continuous pain, pus discharge and swelling around the tooth and gums are classified as mild, moderate or severe. For these variables, experts employ Gaussian membership functions. Assume that z is the variable in fuzzy membership functions, c is the centre of the fuzzy set and σ is the width of the fuzzy set. Eqs. (5-7) provide Gaussian membership functions for continuous pain, while Eqs. (8-10) provide Gaussian membership functions for pus discharge and swelling around the tooth and gums.

Mastication pain, sensitivity to hot or cold, reversible and irreversible pulpitis

Sensitivity to hot and cold can be classified into extreme hot, mild hot, normal cold and extreme cold. Mastication pain, reversible and irreversible pulpitis, on the other hand, can be classified into two linguistic terms, Yes and No. Let z represents the value of mastication pain, sensitivity to hot or cold and reversible and irreversible pulpitis. The fuzzy set for the terms mild hot, normal cold and No is governed by the Z-shaped membership function given in Eq.(11) and the fuzzy set for the terms extreme hot, extreme cold and Yes is governed by the S-shaped membership function given in Eq (12).

Designing fuzzy rules

The proposed system made a decision based on fuzzy if-then rules. Deep cavity, shallow cavity, continuous pain, pus/sinus discharge, swelling around tooth and gums, mastication (chewing) pain, sensitivity to hot and sensitivity to cold are the eight input variables we chose. These parameters have linguistic variables as per [Table 1]. The system as a whole is governed by 665 rules. The following are some of the rules that have been added to the rule base of the fuzzy inference system:

Rule 1: If (deep cavity is low probability) and (shallow cavity is high probability) and (continuous pain is mild) and (pus/sinus discharge is mild) and (swelling around tooth and gums is mild) and (mastication (chewing) pain is no) and (sensitivity to hot is mild hot) and (sensitivity to cold is normal cold) then (reversible pulpitis is yes) and (Irreversible pulpitis is no).

Rule 72: If (deep cavity is medium probability) and (shallow cavity is medium probability) and (continuous pain is mild) and (pus/sinus discharge is mild) and (swelling around tooth and gums is mild) and (mastication (chewing) pain is yes) and (sensitivity to hot is extreme hot) and (sensitivity to cold is extreme cold) then (reversible pulpitis is yes) and (irreversible pulpitis is no).

Rule 384: If (deep cavity is low probability) and (shallow cavity is high probability) and (continuous pain is mild) and (pus/sinus discharge is moderate) and (swelling around tooth and gums is severe) and (mastication (chewing) pain is no) and (sensitivity to hot is mild hot) and (sensitivity to cold is normal cold) then (reversible pulpitis is no) and (irreversible pulpitis is yes).

Rule 663: If (deep cavity is high probability) and (shallow cavity is low probability) and (continuous pain is severe) and (pus/sinus discharge is severe) and (swelling around tooth and gums is severe) and (mastication (chewing) pain is no) and (sensitivity to hot is extreme hot) and (sensitivity to cold is normal cold) then (reversible pulpitis is no) and (irreversible pulpitis is yes).

Fuzzy rule-based inference mechanism and defuzzification

The most widely used fuzzy rule-based models are the Mamdani and Takagi–Sugeno–Kang models. We chose the Mamdani because it can handle fuzzy sets in the antecedent and consequent parts of if-then rules, as we used fuzzy variables for both input and output in our rule-base. The eight antecedent variables and two consequent variables used in this study are shown in [Table 1].

The system output was defuzzified using the Centroid method in order to provide the dentist with a single predictive value for each class of pulpitis. The defuzzified output Zcentroid of centroid method is given in Eq.(13):

To determine the treatment, we look at Zcentroid. If Zcentroid > 0.5, the result is Yes; otherwise, the result is No. This is applied to all the class of pulpitis.

Results

This section focuses on the CNN-fuzzy model's outcome. First, the proposed CNN models' results were focused on, and then the CNN-fuzzy models' diagnosis results were compared to the diagnosis results of a group of doctors.

[Figure 1] presents the proposed CNN architecture for classifying dental cavities as deep or shallow cavities. Initially, the efficiency of this architecture is evaluated by comparing it to that of the VGG-16, VGG-19 and Xception models, which are fully addressed in the discussion section. This architecture achieved an accuracy of 85% [Figure 2]a and [Figure 2]d, which is approximately 11% higher than the traditional VGG16, VGG19 and Xception approaches; as a result, this model is appropriate for CNN-fuzzy approaches. Due to the limited dataset, we tested this model with 50 images of deep cavities and 50 images of shallow cavities. [Figure 3]a represents the predicted probability for a deep cavity, whereas [Figure 3]b represents the predicted probability for a shallow cavity as a sample of 100 images from the testing dataset.

Figure 2: Results of (a) Proposed CNN, (b) CNN-fuzzy approach, (c) expert prediction and (d) it's comparison. CNN: Convolutional neural network

Click here to view

Figure 3: Result of proposed CNN approach; (a) predicted probability for deep cavity, (b) predicted probability for shallow cavity. CNN: Convolutional neural network.

Click here to view

Our precise and reliable proposed CNN model predicts the probability of a deep cavity and a shallow cavity, which is then passed to a fuzzy model, which combines the predicted probability of a deep cavity and a shallow cavity with the symptoms listed in [Table 1] to determine whether the pulpitis is reversible or irreversible. To diagnose reversible or irreversible pulpitis, we used a completely new dataset of 100 images, 50 of which were of irreversible pulpitis and 50 of which were of reversible pulpitis, and it was unknown to both our proposed model and the group of five dentists. [Figure 2]b represents the results of the proposed CNN-fuzzy model for diagnosing reversible or irreversible pulpitis and [Figure 2]c represents the results of a group of doctors in terms of a confusion matrix. [Figure 2]d characterises the confusion matrixes of both by correlating the quantified results of sensitivity, specificity, precision, accuracy, f1 score and Matthews correlation coefficient. The CNN-fuzzy approach has a 94% accuracy, which is 7% greater than expert prediction. The highly accurate novel CNN-fuzzy approach has demonstrated its potential for computer-assisted diagnosis of reversible or irreversible Pulpitis.

Discussion

Our research question is how to accurately diagnose reversible or irreversible pulpitis because radiographs have extremely small length differences, making depth-wise dental pulpitis classification complicated and the application of traditional Boolean logic in human reasoning may result in an inability to evaluate the degree of severity of a symptom and radiographs when trying to diagnose dental pulpitis. This is due to the imprecision and ambiguity inherent in dental diagnostic procedures. In order to respond, we used the CNN-fuzzy approach, which offers a mathematical approach for representing such imprecise information for pulpitis diagnosis by correlating symptoms with radiographs.

We require a robust classifier due to extremely small length differences in cavity radiographs. [Figure 1] presents a highly tuned CNN-based cavity classifier. We used two approaches to tune the hyperparameters of the proposed CNN: Visualisation of Feature Maps and Gradient-Weighted Class Activation Map (Grad-CAM). The feature maps, also known as activation maps, keep track of the data used with filters such as image pixels or other feature maps. The primary goal of visualising a feature map for specific source images is to determine which attributes in the feature maps are identified or retained, which aids in tuning CNN filter parameters. [Figure 4]a shows a sample of the model's activation maps for deep cavities and [Figure 4]b shows a sample of activation maps for shallow cavities.

Figure 4: Activation maps of convolutional layer 1 for (a) deep cavities and (b) shallow cavities.

Click here to view

Selvaraju et al.^[27] created Gradient Weighted Class Activation Mapping (Grad-CAM), a method that provides an illustrative view of deep learning methods. Grad-CAM generates a visual summary for any deeply related CNN, assisting in model selection when conducting classification or prediction tasks. The proposed model is used as a detection method on an X-ray image of a dental cavity. Grad-CAM was calculated for class C classification as a weighted sum of all feature maps produced by the CNN's final convolution layer.^[28]

Grad-CAM is defined for class C as Mc, where is the weight gained by calculating the gradient of a prediction score concerning the k^th feature map and f_k (a, b) is the activation at the spatial element (a, b) in the k^th feature map.

Grad-CAM is applied to all layers of the proposed network after the predicted label is quantified using the model. The images in [Figure 5]a and [Figure 5]b are Grad-CAM-based visualisations of deep and shallow cavity images from the proposed model.

Figure 5: Grad-CAM visualisations of (a) deep cavity images and (b) shallow cavity images

Click here to view

Both approaches, Visualisation of Feature Maps and Grad-CAM, aid in fine-tuning the hyperparameters of the CNN model represented in [Figure 1]. After fine-tuning the hyperparameters of the proposed CNN model, we compared the results of our model to the results of standard models such as VGG16, VGG19 and Xception, and the accuracy is shown in [Table 2].

Table 2: Convolutional neural network-based model's accuracy

Click here to view

Our approach achieved an accuracy of 85%, which is approximately 11% greater than traditional VGG16, VGG19 and Xception approaches. Once our CNN predicted the probability of a deep and shallow cavity, we merged it with symptoms associated with reversible or irreversible pulpitis and used it as an input variable in our CNN-fuzzy model, which is ruled by 665 rules. [Figure 6] presents a surface view of all diagnostic rules for dental pulpitis. The CNN-fuzzy approach has an accuracy of 94%, which is 7% greater than an expert prediction, demonstrating its potential for computer-aided diagnosis of reversible or irreversible pulpitis.

Figure 6: Surface view of all CNN-fuzzy model rules for reversible or irreversible pulpitis diagnosis. CNN: Convolutional neural network.

Click here to view

The CNN-fuzzy system's diagnosis is consistent with dentists' predictions. Since the system has been tested properly and the predictions are consistent with those provided by our dentists, we conclude that the system is functioning in a manner similar to the general intelligence of the dentists. This builds confidence that the system is capable of assisting dentists in diagnosing dental pulpitis by reducing false diagnoses. This system offers a 'second opinion' type of service to dentists who use it and may even be able to reduce the current false diagnosis problem in dental practises. In the future, the enormous potential of these CNN-fuzzy approaches may prove to be very promising in designing symptom and radiograph-based decision support systems for other dental diseases.

Conclusion

A CNN-fuzzy-based approach was developed to assist dentists in clinical decision-making. This method can be used as a 'second opinion' by an expert to assist dentists in the diagnosis of reversible and irreversible pulpitis. Dentists frequently struggle to determine whether pulpitis is reversible or irreversible due to the vagueness and imprecise values of the sign–symptom, as well as the extremely small length differences between 'deep cavities' and 'shallow cavities' in radiographs, which makes depth-wise dental cavity classification difficult and leads to pulpitis misdiagnosis. The accuracy of the CNN-fuzzy approach is 94%, which is 7% greater than an expert prediction, proving its potential for computer-aided diagnosis of pulpitis as well as increasing confidence in the dentist's decision-making, leading to a more effective diagnosis of pulpitis.

Acknowledgement

The authors would like to thank Drs. Ankur J Parmar, Mansi Gohil, Dhara Likhiya and Mayur Kasundra for their time and extensive knowledge of dental pulpitis and dental radiology, as well as for sharing their dataset. There are no conflicts of interest to report. This article did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.

References

1.	Rechenberg DK, Galicia JC, Peters OA. Biological markers for pulpal inflammation: A systematic review. PLoS One 2016;11:e0167289.
2.	Albahbah AA, El-Bakry HM, Abd-Elgahany S. Detection of caries in panoramic dental X-ray images using back-propagation neural network. Int J Electron Commun Comput Eng 2016;7:250.
3.	Singh P, Sehgal P. Automated Caries Detection Based on Radon Transformation and DCT. 2017 8^th International Conference on Computing, Communication and Networking Technologies (ICCCNT). IEEE; 2017.
4.	Haghanifar A, Majdabadi MM, Ko SB, Paxnet: Dental caries detection in panoramic X-ray using ensemble transfer learning and capsule classifier. arXiv 2020, arXiv:2012.13666.
5.	Al Kheraif AA, Wahba AA, Fouad H. Detection of dental diseases from radiographic 2d dental image using hybrid graph-cut technique and convolutional neural network. Measurement 2019;146:333-42.
6.	Lakshmi MM, Chitra P. Classification of Dental Cavities from X-Ray Images Using Deep CNN Algorithm. 2020 4^th International Conference on Trends in Electronics and Informatics (ICOEI)(48184). IEEE; 2020.
7.	Lakshmi MM, Chitra P. Tooth Decay Prediction and Classification from X-Ray Images Using Deep CNN. 2020 International Conference on Communication and Signal Processing (ICCSP). IEEE; 2020.
8.	Geetha V, Aprameya KS, Hinduja DM. Dental caries diagnosis in digital radiographs using back-propagation neural network. Health Inf Sci Syst 2020;8:1-14.
9.	Rad AE, Rahim MS, Kolivand H, Norouzi A. Automatic computer-aided caries detection from dental x-ray images using intelligent level set. Multimed Tools Appl 2018;77:28843-62.
10.	Sornam M, Prabhakaran M. A New Linear Adaptive Swarm Intelligence Approach Using Back Propagation Neural Network for Dental Caries Classification. 2017 IEEE International Conference on Power, Control, Signals and Instrumentation Engineering (ICPCSI). IEEE; 2017.
11.	Lee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J Dent 2018;77:106-11.
12.	Singh P, Sehgal P. GV Black dental caries classification and preparation technique using optimal CNN-LSTM classifier. Multimed Tools Appl 2021;80:5255-72.
13.	Koutsouri GD, Berdouses E, Tripoliti EE, Oulis C, Fotiadis DI. Detection of Occlusal Caries Based on Digital Image Processing. 13^th IEEE International Conference on BioInformatics and BioEngineering. IEEE; 2013
14.	Zhang X, Liang Y, Li W, Liu C, Gu D, Sun W, et al. Development and evaluation of deep learning for screening dental caries from oral photographs. Oral Dis 2022;28:173-81.
15.	Sonavane A, Yadav R, Khamparia A. Dental Cavity Classification of Using Convolutional Neural Network. IOP Conference Series: Materials Science and Engineering. Vol. 1022. No. 1. IOP Publishing; 2021.
16.	Srivastava MM, Kumar P, Pradhan L, Varadarajan S. Detection of tooth caries in bitewing radiographs using deep learning. arXiv e-print 2017; arXiv:1711.07312. https://arxiv.org/abs/1711.07312.
17.	Cantu AG, Gehrung S, Krois J, Chaurasia A, Rossi JG, Gaudin R, et al. Detecting caries lesions of different radiographic extension on bitewings using deep learning. J Dent 2020;100:103425.
18.	Schwendicke F, Elhennawy K, Paris S, Friebertshäuser P, Krois J. Deep learning for caries lesion detection in near-infrared light transillumination images: A pilot study. J Dent 2020;92:103260.
19.	Casalegno F, Newton T, Daher R, Abdelaziz M, Lodi-Rizzini A, Schürmann F, et al. Caries detection with near-infrared transillumination using deep learning. J Dent Res 2019;98:1227-33.
20.	Hwang JJ, Jung YH, Cho BH, Heo MS. An overview of deep learning in the field of dentistry. Imaging Sci Dent 2019;49:1-7.
21.	Schwendicke F, Golla T, Dreher M, Krois J. Convolutional neural networks for dental image diagnostics: A scoping review. J Dent 2019;91:103226.
22.	Khanagar SB, Al-Ehaideb A, Maganur PC, Vishwanathaiah S, Patil S, Baeshen HA, et al. Developments, application, and performance of artificial intelligence in dentistry – A systematic review. J Dent Sci 2021;16:508-22.
23.	Skalski P. Gentle Dive into Math behind Convolutional Neural Networks. Available from: https://towardsdatascience.com/gentle-dive-into-math-behind-convolutional-neural-networks-79a07dd44cf9. [Last accessed on 2019 May 05].
24.	Allahverdi N, Torun S, Saritas I. Design of a Fuzzy Expert System for Determination of Coronary Heart Disease Risk. Proceedings of the 2007 International Conference on Computer Systems and Technologies; 2007.
25.	Mago VK, Mago A, Sharma P, Mago J. Fuzzy logic based expert system for the treatment of mobile tooth. Adv Exp Med Biol 2011;696:607-14.
26.	Mago VK, Bhatia N, Bhatia A, Mago A. Clinical decision support system for dental treatment. J Comput Sci 2012;3:254-61.
27.	Selvaraju RR, Cogswell M, Das A, Vedantam R, Parikh D, Batra D. Grad-Cam: Visual Explanations from Deep Networks via Gradient-Based Localization. Proceedings of the IEEE International Conference on Computer Vision; 2017.
28.	Kim I, Rajaraman S, Antani S. Visual interpretation of convolutional neural network predictions in classifying medical image modalities. Diagnostics 2019;9:38.