PROJECT REPORT

On
“DENTAL CARIES DETECTION USING CONVOLUTONAL
NEURAL NETWORK(CNN)”
Submitted In Partial Fulfilment of the Requirements
For
V SEMESTER
BACHELOR OF COMPUTER APPLICATIONS
SUBMITTED BY
HEMA S S U18EE21S0008
SWATHI M S U18EE21S0130
ANJANA A M U18EE21SO221

Under the guidance of

Prof. Mamatha. M, B.E, MTech. Prof. Shruthi M G, MCA,(Ph.D)
Assistant Professor Assistant Professor

DEPARTMENT OF COMPUTER SCIENCE

MAHARANI LAKSHMI AMMANNI COLLEGE FOR WOMEN
AUTONOMOUS
Science P.O, Malleshwaram 18th Cross, Bengaluru, Karnataka 560012

1
DENTAL CARIES INTRODUCTION
DETECTION USING Many people often overlook dental care.
Without going to the dentist, detecting any
CONVOLUTIONAL caries or issues related to the teeth is very
intricate. Dental caries concludes to tooth pain
NEURAL and tooth loss. It is significant to detect dental
caries early on, in order to prevent it from
NETWORK(CNN) aggravating. According to WHO (World
Health Organization), dental diseases are
costly to get treated, they take up about 5-10%
of the healthcare budget. The normal
ABSTRACT procedure for the diagnosis of dental caries is
through the use of an x-ray, visual inspection,
a dental probe, and handheld mirrors .
However, an X-ray is found to be harmful and
Dental caries are common chronic should be avoided . Even though exposure to
infectious oral disease affecting most radiation is minimum, continuous exposure
can have a long-term effect by damaging the
teenagers and adults. Previous studies
molecular structure and causing harm. Another
depict that X-rays are often used for issue with the use of X-rays is that they cannot
detecting dental caries. This study wants to detect ‘hidden carries’ (invisible at the surface
put special importance on avoiding the use of teeth) in an accurate manner. In the field of
of X-rays as they cannot detect hidden dentistry, the early detection and diagnosis of
caries. Therefore, it is necessary to detect dental caries (commonly known as cavities)
play a crucial role in preserving oral health and
dental caries in accurate manner with
preventing more severe dental issues. With the
proper tools. advancements in machine learning and
computer vision, it is now possible to leverage
This study will serve the purpose of taking these technologies to develop automated
as a input an image of teeth, being able to systems that aid in the identification and
detect where dental caries are located. The classification of caries in oral images.
primary objective is to enhance early
In this project, we aim to build a machine
diagnosis, we employed a methodology
learning model for the detection of caries in
Convolutional Neural Network(CNN) oral images. We will employ techniques such
architecture trained on diverse dataset as data augmentation and regularization to
comprising high resolution oral images. enhance the model's generalization. By
The dataset comprised roughly 300 images utilizing a dataset containing labelled images
where 260 images are training dataset of oral scans, we will train and evaluate the
which involves both caries and no caries model's performance in accurately classifying
images, remaining 40 images are testing images as either showing signs of caries or
dataset also involves caries and no caries. being caries-free.
In particular ,comparing a model based on
test accuracy score, F-score, and ROC
curve, it appears that CNN gives test
accuracy score of 87%.

2
 LITERATURE SURVEY Glogauer, Alistair E W Johnson, Falk
Schwendicke, Carlos Quinonez shows a
Niharika Bhattacharjee worked on research work on Detecting dental caries
Automated Dental Cavity Detection on oral photographs using artificial
System Using Deep Learning and intelligence,out of 3410 identified records,
Explainable AI, which takes training, 506 19 studies were included with six and
de-identified images from online sources seven studies having low risk of biases and
and consenting human participants were applicability concerns for all the domains,
collected. Using curriculum learning, a respectively. Metrics varied widely and
ResNet-27 architecture proved to be most were assessed on multiple levels. F1-
optimal after achieving 82.8% accuracy scores for classification and detection tasks
and 1.0 in sensitivity. Visual explanations were 68.3%–94.3% and 42.8%–95.4%,
for the system’s diagnoses were also respectively. Irrespective of the task, F1-
generated using Local Interpretable Model scores were 68.3%–95.4% for professional
Agnostic Explanation. This system can cameras, 78.8%–87.6%, for intraoral
explain its diagnoses to users in an cameras, and 42.8%–80% for smartphone
understandable manner, which is a crucial cameras. Limited studies allowed
skill employed by dentists[1] assessing AI performance for lesions of
different severity[3]

Malitha Humayun Kabir worked on

Automated caries detection with
smartphone colour photography using
machine learning , the obtained images Jae-Hong Lee a, Do-Hyung Kim a, Seong-
were evaluated for caries diagnosis with Nyum Jeong a, Seong-Ho Choi b shows a
the ICDAS II codes, and were labelled into research work on Detection and diagnosis
three classes: “No Surface Change” of dental caries using a deep learning-
(NSC); “Visually Non-Cavitated” (VNC); based convolutional neural network
“Cavitated” (C). Then, a two steps algorithm, the aim of their current study
detection scheme using Support Vector was to evaluate the efficacy of deep CNN
Machine (SVM) has been proposed: “C algorithms for detection and diagnosis of
versus (VNC + NSC)” classification, and dental caries where a total of 3000
“VNC versus NSC” classification. The periapical radiographic images were
accuracy, sensitivity, and specificity of divided into a training and validation
best model were 92.37%, 88.1%, and dataset (n = 2400 [80%]) and a test dataset
96.6% for “C versus (VNC + NSC),” (n = 600 [20%]). A pre-trained GoogLeNet
whereas they were 83.33%, 82.2%, and Inception v3 CNN network was used for
66.7% for “VNC versus NSC.” Although preprocessing and transfer learning. The
the proposed SVM system required further diagnostic accuracy, sensitivity,
improvement and verification, with the specificity, positive predictive value,
data only imaged from the smartphone, it negative predictive value, receiver
performed an auspicious potential for operating characteristic (ROC) curve, and
clinical diagnostics with reasonable area under the curve (AUC) were
accuracy and minimal cost[2]. calculated for detection and diagnostic
performance of the deep CNN
algorithm[4].

Mohammad Moharrami, Julie Farmer,

Sonica Singhal, Erin Watson, Michael
3
Farzeen Khan , Muhammad khan, naveed
ullah worked on Dental caries and non
caries detection using deep learning, a
number of Deep Learning algorithms were
implemented, and their performances were
evaluated using recall, precision, F1-score,
and accuracy. In comparison to CNN,
LeNet and AlexNet deep learning
techniques, VGG16 has the best
performance, scoring accuracy of 98.99%,
F1-score of 0.96% with precision of
0.95%, and a recall of 0.97%. This
suggested paper demonstrated that DL is
strongly advised for dental professionals to
use in helping them make decisions for the
early diagnosis and treatment of dental
caries[5].

1. CONVOLUTION NEURAL
NETWORK
(CNN)ARCHITECTURE

fig 5.1.1 The architecture of CNN

4
IMPLEMENTATION

CNN

Performance
analysis
DATASET
Training the
models

Classify
Split data for caries
Processing and training and Test the
augmentation models and no
testing
caries

Fig 2. Proposed methodology using CNN

2.1 DATASET COLLECTION 2.2 DATA PREPROCESSING

The original dataset consists of 300 images in There are several issues in dental images
that 260 training images which includes both dataset ,the data must be clean for models to
caries and no caries images and 40 testing process it. Various types of pre-processing
images, which also includes both caries and no tasks such as image resize and normalization is
caries. required. The image can be resized to get rid
of unnecessary information in the background.
Whole dataset images were divided into train Machine Learning model requires the data
and test groups. 80% of data was used for being trained to have the same size. Therefore,
training, remaining 20% used for testing. an immediate image of resizing is essential to
prepare for Machine Learning models.

2.3 DATA AUGMENTATION

5
Data augmentation is a set of techniques to  Data loading and pre-processing
artificially increase the amount of data by  Model architecture
generating new data points from existing  Model Compilation
data. This includes making small changes  Data Augmentation
to data or using deep learning models to  Model training
generate new data points. Machine learning  Model Evaluation
applications especially in the deep learning  Save the model
domain continue to diversify and increase  Deployment
rapidly. Data-centric approaches to model
development such as data augmentation
techniques can be a good tool against
challenges that the artificial intelligence 2.5 TESTING
world faces. Data augmentation is useful to
improve the performance and outcomes of Testing involves evaluating its
machine learning models by forming new performance on a separate set of data not
and different examples to train datasets. If used during training . Testing data sets
the dataset in a machine learning model is contains 40 images.
rich and sufficient, the model performs
better and more accurately. For machine Steps for testing CNN model are:
learning models, collecting and labelling
data can be exhausting and costly  Load pre-trained model
processes. Transformations in datasets by  Prepare test data
using data augmentation techniques allow  Evaluate the model
companies to reduce these operational  Confusion Matrix
costs. One of the steps in a data model is  Visualize predictions
cleaning data which is necessary for high-  Iterate and Improve
accuracy models. However, if cleaning
reduces the representability of data, then
the model cannot provide good predictions
for real-world inputs. Data augmentation 2.6 FEATURE EXTRACTION
techniques can enable machine learning AND CLASSIFICATION
models to be more robust by creating MODULE
variations that the model may see in the
real world.  *Convolutional Layers:* These layers
are responsible for feature extraction.
They use small squares of input data
(convolutional kernels) to learn local
2.4 TRAINING patterns.
 *Max Pooling Layers:* These layers
Training a CNN for dental caries detection down-sample the spatial dimensions of
involoves several key steps.Below is an the input, reducing the computation in
overview of training process. Among 300 the network and focusing on the most
images, 260 images are divided into important features.
training dataset.

6
 *Flatten Layer:* This layer flattens the RESULT
output from the convolutional layers
into a one-dimensional array,
preparing it for classification.
 *Dense Layers:* These fully
connected layers perform the
classification based on the features
extracted by the convolutional layers.
 *Output Layer:* The output layer has
one neuron with a sigmoid activation
function for binary classification
(caries or no caries). For multi-class
classification, you would adjust the
output neurons and activation function
accordingly.

Once you define and compile the model,

you can use this for both feature extraction
and classification during the training
process. The CNN will learn to Fig 3.1 Bar plot for class distribution in
Train Data
automatically extract relevant features from
dental images and make predictions based
on those features.

Fig. 3.2 Bar plot for distribution in Test

Data

7
Fig 3.3 Training and Validation accuracy

8
Fig 3.4 Training and Validation loss

CONCLUSION

A Convolution Neural Network(CNN)

model is used for detecting and classifying

9
caries and no caries images using machine Advances in deep learning and image
learning . In particular ,comparing a model processing can lead to improved accuracy
based on test accuracy score, F-score, and and efficiency in identifying dental caries.
ROC curve, it appears that CNN gives test
Integration with emerging technologies
accuracy score of 87%.
like augmented reality and telemedicine
In this project, we aim to build a machine could enhance remote diagnostics and
learning model for the detection of caries preventive care.
in oral images. We will employ techniques
Collaborations between dental
such as data augmentation and
professionals and AI experts can further
regularization to enhance the model's
refine these systems, contributing to more
generalization. By utilizing a dataset
precise and accessible oral health
containing labeled images of oral scans,
assessment.
we will train and evaluate the model's
performance in accurately classifying Ongoing research and development in this
images as either showing signs of caries or field may lead to the creation of user-
being caries-free. friendly tools for early detection and
personalized treatment planning
The successful development of an
automated caries detection system holds
immense potential for improving dental
healthcare practices. Through the
implementation of a convolutional neural
network and the application of data
augmentation and regularization
techniques, we have demonstrated the REFERENCES
ability to effectively classify oral images
and identify the presence of dental caries.
Our model achieved impressive accuracy [1]. J.D. Featherstone The science and
in distinguishing between caries and non- practice of caries prevention J. Am. Dent.
caries images, showcasing the power of Assoc.(2000).
machine learning in healthcare.
[2]. Amrollahi et al. Recent advancements
in regenerative dentistry: a review Mater.
Sci. Eng. C: Mater. Biol. Appl. (2016)

[3]. P.S. Stein et al. Tooth loss, dementia

and neuropathology in the Nun study J.
Am. Dent. Assoc.(2007)
FUTURE SCOPE
[4]. M.J. Tyas et al. Minimal intervention
The future scope for dental caries detection
dentistry—a review. FDI commission
using Convolutional Neural Networks
project 1–97 Int. Dent. J.(2000)
(CNN) in oral images holds great
potential.

10
[5]. R.H. Selwitz et al. Dental
cariesLancet(2007)
[13]. Pitts, N. J. D. R. The use of bitewing
[6]. S. Matalon et al. Diagnosis of radiographs in the management of dental
approximal caries: bite-wing radiology caries: scientific and practical
versus the ultrasound caries detector. An in considerations. Dentomaxillofac.
vitro study Oral Surg. Oral Med. Oral Radiol. 25, 5–16 (1996).
Pathol. Oral Radiol. Endod.(2003)

[7]. D.C. Gakenheimer The efficacy of a

computerized caries detector in intraoral [14]. Braga, M. M., Mendes, F. M. &
digital radiography J. Am. Dent. Assoc. Ekstrand, K. R. J. D. C. Detection activity
(2002) assessment and diagnosis of dental caries
lesions. Dent. Clin. 54, 479–493 (2010).
[8]. E.D. Beltran-Aguilar et al.
Surveillance for dental caries, dental
sealants, tooth retention, edentulism, and [15]. Pretty, I. A. & Ellwood, R. P. The
enamel fluorosis—United States, 1988– caries continuum: opportunities to detect,
1994 and 1999–2002 MMWR Surveill. treat and monitor the re-mineralization of
Summ(2005) early caries lesions. J. Dent. 41(Suppl 2),
S12-21 (2013).
[9]. Centers for Disease Control and
Prevention, National Health and Nutrition
Examination Survey (NHANES) 1999–
2002

[10]. R.G. Watt et al. Tooth loss and

cardiovascular disease mortality risk—
results from the Scottish Health Survey
PLoS One(2012)

[11]. Young, D. A. et al. The American

Dental Association Caries Classification
System for clinical practice: A report of
the American Dental Association Council
on Scientific Affairs. J. Am. Dent.
Assoc. 146, 79–86 (2015).

[12]. Hilton, T. J., Ferracane, J. L. &

Broome, J. C. Summitt’s Fundamentals of
Operative Dentistry: A Contemporary
Approach (Quintessence Publishing
Company Incorporated, 2013).

11