A Comparative Study on Tongue Tumor Detection and Classification

Using Neuro Dynamic Ensemble Fusion Classifier

S. Preethi

, K. Raju

, Rajalakshmi D

, P. Venkata Subba Reddy

, Shaik Taher

and Sannapu Deepak

Department of Information Technology, E. G. S. Pillay Engineering College, Nagapattinam, Tamil Nadu, India

Department of Artificial Intelligence and Machine Learning, Rajalakshmi Engineering College, Chennai, Tamil Nadu, India

Department of Computer Science and Engineering, R.M.D Engineering College, Chennai, Tamil Nadu, India

Keywords: Image Processing, Deep Learning, Tongue Tumor, Oral Cancer, Neuro Dynamic Ensemble Fusion.

Abstract: Recently, the deep learning technique plays a vital role in health care industries to classify and diagnose any

disease at early stage. By applying deep learning algorithms, the diseases such as heart disease, brain tumor,

lung disorder, and several other deadly diseases has been early diagnosed. This promising technique endlessly

extends its viability in detecting oral cancer. This life threating tumor can forms on any region of the mouth

namely uvula, tonsil, gum, palate and cheeks. As mouth has been the crucial part in our body, because it

performing functions such as eating, sleeping and breathing. Hence, early detection and diagnosing of oral

cancer is mandatory. If it is not treated, this disease endangers the life of humans. Earlier the traditional

convolutional neural networks were used for diagnosing oral cancer. But the advancement of information

technology brings development in neural networks. This study proposes a novel Neuro dynamic ensemble

fusion (NDEF) classifier to enhance the detection of oral cancer at earliest. The proposed model is tested on

the publically accessible oral cancer dataset and comparing its performance with other classifiers including a

hybrid RCNN and ResNet—50, VGG16, and U-Net. This proposed classifier exhibits higher accuracy of

96.27%, precision and recall of 96% and 94%, respectively. The NDEF has obtained promising results and

accurately detected the affected regions as well.

1 INTRODUCTION

The term “cancer” is known as a disease where the

abnormal body cells dissect enormously and damage

the tissues of the whole body. About 200 more types

of cancers have been discovered. The most frequently

occurring cancers take place in the breast, lungs, skin,

stomach, liver and tongue. According to the World

Health Organization (WHO), oral cancer is one of the

most dominant and widely spreading types of cancer

and its mortality rates are also high in several

countries notably in South Asia. Usage of tobacco,

consuming heavy amounts of alcohol, and HPV

infection with some specific genetic factor cause the

tongue tumor in which the cancerous cells dissociate

from the tumor and spread to the other parts of the

body especially inside the mouth, head, neck, in the

areas of lungs and the areas which are close to the

lymph nodes. In India, nearly 52,000 people die of

tongue tumor per year. So, detecting and curing it in

its early stage is mandatory which saves thousands of

people from its threat.

In human life, the Artificial Intelligence (AI)

techniques play a vital role which helping people

being endangered. Deep learning is one among them

that has a significant impact on detecting tongue

tumor. Some of the earliest detection methods for

tongue tumor were, (K. Nakanura et al., 2012) used

Raman Spectroscopy-based system for detecting oral

cancer which is non-invasive and reduses the risk of

discomfort but failed to produce larger number of

datasets and could be more expensive. And then, (S.

Wang et al., 2017) detected tongue tumor with the

help of CNN (Convolutional Neural Network) based

detection with simple architecture along with the

robust feature extraction it couldn’t provide a larger

number of datasets. Following this, (y. zhang et al.,

2018) used Auto encoder-based detection which

provided unsupervised learning but only had limited

features. Additionally, (S.S Iyer et al., 2019) used

880

Preethi, S., Raju, K., D., R., Reddy, P. V. S., Taher, S. and Deepak, S.

A Comparative Study on Tongue Tumor Detection and Classiﬁcation Using Neuro Dynamic Ensemble Fusion Classiﬁer.

DOI: 10.5220/0013875000004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

880-887

ISBN: 978-989-758-777-1

transfer learning-based detection which is an efficient

training model but it ended up with a limited fine

issue. The above researchers were some of the earliest

detection of tongue tumor using the existing models.

This study let us analyze how to recognize and

classify tongue tumor with proper pre-processing

filtering called Multi Scale Adaptive Filtering by

image resizing, and denoising in addition to some of

the feature extraction processes which is followed by

the specific segmentation called Fractal Texture

Mapping for Neurodegenerative Segmentation

(FTM-NS) technique for depicting the affected areas.

Finally, the process is classified using the

classification model named as Neuro-Dynamic

Ensemble Fusion to detect and eradicate the tumor in

its early stage. This study also serves as a comparative

analysis of existing and proposed tongue tumor

detection techniques for better results.

2 RELATED WORKS

To utilize deep learning for detecting the abnormal

growth of oral tissue (Welikala et al., 2021) applied

an Artificial Neural Network (ANN) for the

automated detection of oral lesions. This study

promotes the early identification of oral lesions which

can significantly reduce treatment costs and even

prevent mortality rates. To classify the image, the

ResNet-101 model was employed achieving an

accuracy of 87.07%. Furthermore, the damaged

tissues in the images were accurately identified with

78.3% of precision. Though the performance in

identifying followed by classifying tongue tumours

using DNN was demonstrated acceptably, several

limitations were also noted, which include limited

data set size inconsistent annotations lack of external

validation restricted evaluation metrics followed by

the absence of comparative analysis with limited

clinical validation. (Nandita et al., 2022) employed

both deep learning and machine learning techniques,

which promote the identification of tongue tumours.

In this study, a Convolutional Neural Network (CNN)

with 43 deep layers was engaged to predict the data.

This results in detecting CT scan images with high

accuracy which is also effectively stimulating

malignant oral lesions with utmost precision. AI has

become apparent in the diagnosis of several diseases,

including cancer. To identify the tongue lesions,

(Panigrahi et al., 2022) employed histopathological

images. This study assessed 6 widely used algorithms

called Support Vector Machine (SVM), Random

Forest, Neural Network, Simple Bayes, Decision tree

and K- nearest neighbor (KNN) which are the most

relevant methods in classifying oral lessons.

Additionally, the study admitted that the neural

network algorithm achieved its reasonable accuracy

of 90.4% with satisfactory potential in diagnosing the

disease. (Singh et al., 2022) introduced an innovative

intelligent computing framework for deducting

tongue tumours. He evaluated the strategy with the

help of the disease imaging data. This concluded in

revealing the tumor in their early stage. To distinguish

healthy tissue from cancerous tissue (Jeng et al.,

2022) utilised Raman spectroscopy through specific

subsite analysis. This focused on the tongue, gingival

and buccal mucosa. The classification of healthy and

cancerous tissues was successful by employing

Linear Discriminant Analysis (LDA) followed by

Quadratic Discriminate Analysis in cooperation with

Principle Quality Analysis (PQA). Principally,

Raman's Spectroscopy highlighted the potential in

detecting oral cancer by finding that the proteins,

amino acids and beta carotene served as consequent

biomolecular markers to get rid of cancer. (Sahu et

al., 2023) achieved a sensitivity of 64% and

specificity of 80% with the application of

the Principle Component Liner Discriminate

Analysis Mode which examined the potential of

serum Raman Spectroscopy in diagnosing tongue

tumor. Though they tend to have some limitations,

they lead to optimal performance in spectral data

classification. Despite this, deep learning models

enable automatic feature extraction from raw data to

an end-to-end learning approach. Hence these deep

learning AI models have an optimistic perspective in

improving the accuracy of tumor classification.

3 METHODOLOGY

This section provides a detailed explanation of the

steps followed in proposed technique which includes

dataset collection, pre-processing, segmentation and

classification.

3.1 Dataset Collection

The current study utilizes the oral cancer images

acquired from a publicly available oral cancer data

set. The images of the oral cancer obtained from the

database are in the JPEG format, which is with a

specific resolution of 256 × 256 pixels. The obtained

dataset holds the collection of tongue Figure 1 which

are grouped into two categories, namely cancerous

and non-cancerous images. Furthermore, the images

in the dataset comprise 500 sets of oral cancer images

and 450 sets of non-cancer oral images, which are

A Comparative Study on Tongue Tumor Detection and Classiﬁcation Using Neuro Dynamic Ensemble Fusion Classiﬁer

881

being compiled that can be used for various medical

visualization in the detection and therapeutic

treatment of oral cancer earlier.

Figure 1: Non-cancerous and cancerous tongue.

3.2 Pre-Processing

In this study, the images will be compiled from

classified reports, which offer significant data for

better study. Moreover, the images which are

obtained from the specific source have some

complications which we have to work on like the

quality of the image, and its structure followed by

some denoising process. Though some of the images

are quite clear and well-organized others may be

composed by the presence of noise which makes the

study to be decreased in its accuracy. To overcome

the issue, some of the preprocessing techniques are

essential to enhance the quality of the image,

ultimately promoting the betterment of the proposed

study. For enhancing the image quality effectively,

multi scale adaptive filtering serves as a crucial pre-

processing step in tongue tumor detection.

Denoising: It is very difficult to detect the tongue

tumor accurately from the tissue surrounding the

tumor. It's been more complicated when the image

quality is compromised by noise or even lightning.

This is effectively eradicated with the help of the

multiple scales adaptive efficiently and reducing the

noise by preserving some crucial details, including

the edges of the tumor and contours, therefore making

tumor identification much easier. At first, the noisy

images cannot be detected accurately. So, the

adaptive filter helps in locating the edges of the

images which are composed of noises. Next, the

images which are comprised of noises are

decomposed with the wavelets. Then the edges are

located with the adaptive filters. The edges and the

contours were detected at multiple scales.

Furthermore, the adaptive filters store the details of

the images even after removing the noises. At each

selected scale the process of filtering is applied to the

image. By analysing and adjusting its guidelines

based on the texture, contrast and intensity in that

particular region, the filter is being operated. After the

filtering has been done in multiple scales, the scale

allows the preservation of the edges of small tumours

and the structure of the borders. The processed final

image is then claimed and detailed with enhanced

features making it suitable for further processing.

This Multi-scale adaptive filtering process ensures

the improvement in tumor detection's accuracy, better

edge detection, and adaptability rather to the other

filters. Finally, multi-scale adaptive filtering makes

the tumor detection more reliable by playing a crucial

role in improving image quality.

Figure 2: Original and the filtered image using Multi-scale

adaptive filter.

To remove noise from an image while keeping

important details like edges, a technique called Multi-

scale Adaptive Filtering is used. This method

involves applying filters at different scales to the

image. Each scale reduces noise in a specific way.

This process Figure 2 Shows the Original and the

filtered image using Multi-scale adaptive filter. can

be broken down into mathematical steps to achieve

effective denoising. The general mathematical

equation for the denoising process using Multi-scale

Adaptive filter is given below.

Let I(x,y) be the pixel coordinates of the noisy

image. Initially, the image is decomposed into

Multiple Scales.

Let Is(x,y) represent the image at the s’th scale,

where s =1,2,…, S and S is the number of scales. At

scale s, the adaptive filter Fs is applied to the image

Is(x,y) yielding the filtered image Îs(x,y): Îs (x,y) =

Fs(Is(x,y),σs)

Where σs represent the noise variance that

depends on the scale. The images at different scales

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

882

are then combined to produce the final denoised

output. This is given by, Î (x,y) = ∑Ss=1 ws Îs(x,y)

where ws is the weight assigned to each scale that

depends on factors like the scale or the effectiveness

of the denoising at that scale. Finally, the denoised

image Î (x,y) is used as the input for the detection of

tongue tumor. Hence this ensures the reduction of

noise effectively at multiple scales which allows for

better tongue tumor detection.

3.3 Segmentation

After the Multi-scale adaptive filtering, the enhanced

image is now ready for the tumor

segmentation. Fractal Texture Mapping for

Neurodegenerative Segmentation (FTM-NS) works

by segmenting the texture of the images which consist

tumor. The affected patterns may be fractal which

remains the same across different scales. Hence,

fractal texture mapping helps in segmenting the tumor

of neurodegenerative diseases, which exhibit

complex structures at different scales. It highlights

the irregularities in tissue patterns caused by tumor by

mapping these fractal characteristics. Firstly, it

identifies the affected areas. The affected areas may

have irregular patterns. The fractal method helps in

analyzing the irregularities and detects the changes.

Next, the detected changes of the neurodegenerative

diseases are mapped, which helps in differentiating

the tumorous and the non-tumorous tongue. Finally,

the segmentation process takes place by locating the

affected areas at various scales. Thus, the affected

tissues with complex structures are detected by fractal

geometry, which results in mapping the changes in

the texture for more accurate segmentation in

neurodegenerative diseases. This method enables the

exact segmentation and the reduction of

abnormalities accurately. For a better understanding,

the tumor in a tongue is segmented and the image is

projected below Figure 3.

Figure 3: Segmented tongue using FTM-NS.

3.4 Classification

Neuro-Dynamic Ensemble Fusion (NDEF)

classification is one of the best classifiers for

detecting tongue tumor. It combines various tumor

detecting techniques to classify the tumor in the

tongue with an increased amount of accuracy. It

works by involving the neural networks, where the

information is being processed. These neural

networks can handle the complex patterns which are

in the acquired images. Then the models which

depend on one another are ensembled. Then those

ensemble models work together resulting in

classifying the tumor with accurate prediction.

Finally, the tumor are classified after sorting the data

in the medical images. In brief, this proposed NDEF

combines different neural networks with various

models and fuses the techniques of the models with

the networks to classify the tongue tumor more

accurately. Thus, the Neuro Dynamic Ensemble

Fusion (NDEF) model results in excellent tongue

tumor detection. It outperforms the existing models

like RCNN, ResNet-50, VGG16, and U-Net. NDEF

combines multiple algorithms which increases the

accuracy and robustness. It adapts to complex

patterns and features in medical imaging. This makes

NDEF to be the most effective for tongue tumor

detection. For the earliest detection and diagnosis,

accurate detection of the disease is a must. Several

deep learning models, including RCNN, ResNet-50,

VGG16, and U-Net, are also being used for medical

imaging. Each of the models has its strengths and

weaknesses. This study also compares the

performance of the existing models with the proposed

Neuro Dynamic Ensemble Fusion (NDEF) model for

enhanced tongue tumor detection.

RCNN is mainly used for the detection of objects.

It works by using a CNN to classify every region and

divide the images into regions of interest (RoIs). This

process makes RCNN detect specific regions such as

tumor and the affected areas effectively. Moreover,

RCNN plays a significant role in detecting and

segmenting the tumor within the image. Though it

detects the tumor efficiently it also has some

limitations. One of the major drawbacks is, that the

speed of the image processing is extremely slow

which further processing steps like, feature

extraction, and classification. This also makes the

RCNN perform very slowly in medical diagnostics.

The next major drawback is, that RCNN requires a

larger number of computational resources,

particularly in handling huge medical images. This

can’t be resolved making RCNN work slowly. Hence

in the detection of tongue tumor, RCNN supports

detecting the tumor regions effectively. However, its

slow processing speed limits its work in the earliest

disease diagnostics, as the fastest detection and

diagnosis are very important.

A Comparative Study on Tongue Tumor Detection and Classiﬁcation Using Neuro Dynamic Ensemble Fusion Classiﬁer

883

ResNet-50 is also a deep learning, convolution

neural network model. It uses residual connections

which makes deeper networks. It is very significant in

detecting even a tumor with minor characteristics

which is very effective in medical image analysis. Its

enhanced architecture helps us to analyze the tumor

which has irregular features and patterns. It has

residual connections which prevent degradation. This

makes it applicable to deeper networks. ResNet-50 is

known for its larger number of datasets making it

beneficial for analyzing medical images. However,

this classifier too has some limitations. As the model

is very large, the highest memory storage is required.

The next major drawback is, that to gain the highest

accuracy in tongue tumor detection; the model has to

be tuned finely along with the labelled data.

VGG16 is also a deep learning-based convolution

neural network. It has enhanced features which are

well known for its simple architecture and peculiar

structure. The model has 16 layers, with 3x3 filters

which can be applied through all the layers along with

small receptive fields. This makes the medical

diagnosis very easy as it classifies the image

accurately which promotes the model to be popular

among the researchers. Furthermore, its unique

structure helps it to understand and implement the

image classification process very easily. The 16

layers in the model allow for projecting even the

complicated details in the image which makes the

tumor detection very easier. Though it works very

efficiently, this too has some limitations. As the

structure is deep, the model requires the highest

amount of memory storage and increases the

computational costs which results in overfitting.

Additionally, the model gives only a limited number

of datasets. However, VGG16 is very useful in

classifying the tongue tumor but the model may not

work very efficiently due to its overfitting property

and limited number of datasets which results in giving

poor quality data.

U-Net is a popular architecture which used as a

deep learning model for semantic segmentation

mainly in medical image analysis. It works with the

help of an encoder-decoder architecture which has

skip connections that help preserve tumor based

information. This feature makes U-Net perform

accurately in segmenting and classifying the images.

It works well not only in classifying the images but

also in segmenting the tumor edges perfectly in the

detection of tongue tumor. U-Net models are trained

by using a minimum number of datasets with limited

data. Moreover, U-Net has also struggled with some

limitations. As it is very good at segmenting the

images, it couldn’t perform well in classifying the

images. As a result, U-Net requires additional

classification steps in the detection of tongue tumor

to make the treatment quick and easier.

Neuro-Dynamic Ensemble Fusion (NDEF)

classification is one of the best classifiers for

detecting tongue tumor. It combines various tumor-

detecting techniques to classify the tumor in the

tongue with an increased amount of accuracy. It

works by combining multiple neural networks to

improve classification accuracy. Based on the

features obtained from the tumor, their learning

patterns and the data that is imported into the image,

the neural networks operate and update the

classification efficiently. It combines various types of

classifiers, which are from different neural networks

to generate a final decision by improving the

robustness. The combination of the classifiers helps

in predicting the image data from different models

which results in producing the accurate classification.

The following Figure 4 grouped below shows the

classification of the tumor in the tongue and the non-

tumorous tongue for more clarification.

Figure 4: Classification result.

4 RESULT & DISCUSSION

The Proposed system was validated using the

publically available data set which contains Images of

the oral cancer with 256 x 256 pixels that can be used

for medical visualization with correct resolution. For

enhancing the image quality, denoising and filtering

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

884

is done by using the Multi-scale Adaptive Filter and

the proper image is obtained which makes it suitable

for the further processing like segmentation. The

tumor affected area is segmented perfectly with the

help of Fractal Texture Mapping Neurodegenerative

Segmentation. It highlights the irregularities in tissue

patterns caused by tumours by mapping these certain

characteristics. Finally, in classification, the existing

models like RCNN, ResNet50, VGG16 and U-Net are

compared with NDEF's classification model and

concluded that the highest accuracy and reliability is

most effective in tongue tumor deduction is acquired

only by using NDEF classification model.

Further the evaluation metrics namely accuracy,

precision-recall, are widely utilised in the field of

image classification to describe the model’s

performance.

The proposed classifier has achieved the accuracy

of 96.27% more than the other existing models like

RCNN+ResNet50 which have 93.22%, VGG16

which has 90.81% and UNet which has 88.04% of

accuracy which is shown in Figure 5.

Figure 5: Accuracy comparison.

Figure 6: Precision-Recall comparison.

Its effectiveness is also been calculated by the

precision and recall scores in which the proposed

model has achieved the highest precision of 0.96 and

recall score of 0.94 than the existing model which are

discussed in the Figure 6. The training and testing

accuracy obtained by the proposed classifier is

depicted in Figure 7.

Figure 7: Validating the Training and Testing accuracy.

Here, the proposed model has the highest training

and testing accuracy of about 96.27%

For the further enhancement of the proposed

study, the works of different authors using the

existing methods are also compared. From the

proposed NDEF technique, it is known that the

technique has overcome all other methods in

detecting the tongue tumor with the highest accuracy

of 96.27%. In (P. Kalaivani, 2022) the technique uses

the Gabor Filter for increasing image quality along

with the K-Means Clustering segmentation for

detecting affected areas which has an accuracy of

94%. Despite, this the technique used in (L. Li et al.,

2022) that employs Gabor, Sobel, and Median Filters,

for pre-processing and Thresholding, K-Means

Clustering, and Watershed Transform methods for

segmentation achieves an accuracy of 93%. In the

techniques (W. Wang et al., 2023) and (J. Heo, 2022)

the image quality is done by using Multi-Resolution

Analysis Filters and Gaussian filter, Median filter,

CLAHE along with the region-based segmentation

and Mask R-CNN& U-Net segmentation models in

depicting the tumor areas resulting in the accuracies

of 82% and 78.6%, respectively. Technique (T.

Thakuria, 2022) achieves the accuracy of 89.47% by

using Gaussian, Median, CLAHE, Wiener, and

Anisotropic Diffusion Filter and by using FCN,

SegNet, and DeepLab segmenting models. Finally,

NDEF has its significant combination of Multi-scale

Adaptive Filter along with FTM-NS segmentation

technique over all other techniques, which results in

the highest position as the most effective method for

tongue tumor detection. Table 1 Shows the

Comparing the other existing technique with the

proposed technique.

A Comparative Study on Tongue Tumor Detection and Classiﬁcation Using Neuro Dynamic Ensemble Fusion Classiﬁer

885

Table 1: Comparing the other existing technique with the proposed technique.

Author

Name

Dataset Filter Segmentation Accuracy

W. Wang et

al.,

Oral cancer

dataset

Multi-Resolution

Analysis Filters

Region based segmentation 82%

P. Kalaivani

Oral

Histopathology

Dataset

Gabor Filter K-Means Clustering 94%

L. Li et al.,

Oral Cancer

Dataset

Gabor, sobel &

median filters

Thresholding,

K-Means clustering &

Watershed transform

93%

J. Heo

TCEED (Tongue

Cancer

Endoscopic

Dataset)

Gaussian filter,

Median filter&

CLAHE

Mask R-CNN& U-Net 78.6%

T. Thakuria

Oral Cancer

Dataset

Gaussian,

Median, CLAHE,

Wiener and

Anisotropic

diffusion filte

FCN, SegNet & DeepLab 89.47%

Proposed

NDEF

Oral Cancer

Dataset

Multi-scale

Adaptive Filte

FTM-NS 96.27%

5 CONCLUSIONS

The experimental results of the proposed Neuro

Dynamic Ensemble Fusion mechanism have

achieved its maximum accuracy. While comparing

the proposed with the other classification models, the

NDEF approach achieves accurate tumor detection

even at low resolutions. The use of Fractal Texture

Mapping for neurodegenerative segmentation

enhances the exact tumor-affected regions by its

effective segmenting technique and the Multi-scale

Adaptive Filtering helps in eliminating the impurities

in the images including the edges. Thus, the model

helps the information to be preserved on the original

image. The increased prediction level trains the

model to extract the essential features in the feature

extraction technique. As a result, the model achieves

an overall accuracy of 96.27%, which serves as the

perfect replacer of other classification models. The

proposed model's effectiveness is proved by the

evaluation metrics such as precision and recall. In

summary, the proposed model offers efficient and

accurate tongue tumor detection, with a large number

of datasets and decreased detection time. Thus, it

serves as the best alternative approach to any other

models for effective tongue tumor detection.

REFERENCES

K. Nakamura, "Raman spectroscopy for detecting oral

cancer," Diagnostics, vol. 12, no. 12, pp. 2896, 2012.

S. Wang et al., "Detection of tongue tumor using

convolutional neural network (CNN)," Computer

Methods and Programs in Biomedicine, vol. 145, pp.

73-82, 2017, doi: 10.1016/j.cmpb.2017.04.011.

Y. Zhang et al., "Automatic detection of tongue tumor using

auto encoder and support vector machine," IEEE

Access, vol. 6, pp. 38601-38609, 2018, doi:

10.1109/ACCESS.2018.2854823.

S. S. Iyer et al., "Transfer learning-based tongue tumor

detection using deep neural networks," IEEE Access,

vol. 7, pp. 103941-103953, 2019.

doi: 10.1109/ACCESS.2019.2931953

X. Cai, "Breast cancer diagnosis by convolutional neural

network and advanced thermal exchange optimization

algorithm," Computational Mathematics and Methods

in Medicine, vol. 2021, 2021.

Nanditha, B., Geetha Kiran, A. Sanathkumar, M. P., "Oral

cancer detection using machine learning and deep

learning techniques," International Journal of Current

Research and Review, vol. 14, no. 01, pp. 64, 2022.

S. Panigrahi, B. S. Nanda, and T. Swarnkar, "Comparative

analysis of machine learning algorithms for

histopathological images of oral cancer," in Advances

in Distributed Computing and Machine Learning,

Springer, 2022, pp. 318-327.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

886

A. Singh, A. Sahu, and S. Verma, "Computer intelligence

in detection of malignant or premalignant oral lesions:

The story so far," Computer Intelligence in Oncology,

vol. 1016, pp. 187-200, 2022.

C.I. Faur et al., "Raman spectroscopy in oral cavity and

oropharyngeal cancer: a systematic review," Internati-

onal Journal of Oral and Maxillofacial Surgery, vol.,

pp., 2022.

"Deep multi-feature fusion residual network for oral

squamous cell carcinoma classification and its

intelligent system using Raman spectroscopy,"

Biomedical Signal Processing and Control, vol., pp.,

2023.

W. Wang, Y. Liu, and J. Wu, "Early diagnosis of oral

cancer using a hybrid arrangement of deep belief

network and combined group teaching algorithm,"

Scientific Reports, vol. 13, pp. 22073, 2023.

P. Kalaivani, "Oral cancer detection using deep learning,"

Journal of Intelligent Information Systems, vol. 60, no.

2, pp. 347-362, 2022. doi: 10.1007/s10844-021-00761-

L. Li, C. Pu, J. Tao, L. Zhu, S. Hu, B. Qiao, L. Xing, B.

Wei, C. Shi, P. Chen, and H. Zhang, "Development of

an oral cancer detection system through deep learning,"

Computers in Biology and Medicine, vol.

142, pp. 105512, 2022.

doi: 10.1016/j.compbiomed.2022.105512.

J. Heo, "Deep Learning Model for Tongue Cancer

Diagnosis Using Endoscopic Images," Journal of

Clinical Medicine, vol. 11, no. 11, pp. 3024, 2022. doi:

10.3390/jcm11113024.

T. Thakuria, "Deep learning for early diagnosis of oral

cancer via smartphone and DSLR image analysis: A

systematic review," Journal of Oral Pathology &

Medicine, vol. 51, no. 9, pp. 641-653, 2022. doi:

10.1111/jop.13441.

A Comparative Study on Tongue Tumor Detection and Classiﬁcation Using Neuro Dynamic Ensemble Fusion Classiﬁer

887