Pancreatic Cancer Detection Using Deep Learning

Surasani Akhila, G. N. Swamy, Sudabattula Sahithi, Shaik Faheem and Jogi Rajesh

Department of Electronics and Instrumentation Engineering, VR Siddhartha Engineering College, Vijayawada, Andhra

Pradesh, India

Keywords: Pancreatic Cancer, Deep Learning, Convolutional Neural Networks (CNN), MATLAB, Transfer Learning,

Image Segmentation.

Abstract: Pancreatic cancer the most lethal malignancy remains difficult to detect in early stages due to a lack of specific

symptoms in unique tumor morphology. Deep learning, specifically with convolutional neural networks

(CNNs), has demonstrated potential in increasing diagnostic accuracy and facilitating early detection in

medical imaging. The aim of this research is to implement Deep learning algorithms for the detection of

pancreatic cancer using MATLAB. It also illustrates how transfer learning and multimodal image fusion leads

to greater improvement over the proposed model, particularly in scenarios with limited data. MATLAB's

Deep Learning Toolbox and Image Processing Toolbox are used to organize the processing of the images, the

extraction of the features, and the training of the models.

1 INTRODUCTION

One such form is pancreatic cancer, one of the most

aggressive and deadly cancers, with rapid progression

and late diagnosis. Although improvements have

been made in diagnostic imaging technologies, the

prognosis for this cancer is still very poor as

pancreatic tumors are detected too late. Diagnostic

imaging is commonly performed using conventional

imaging (CT, MRI, and endoscopic ultrasound). But

of these approaches, most invariably fail to

Detecting and differentiating cancerous from non-

cancerous tissues, especially because the symptoms

are less prominent or absent during the early cancer

stage. This presents a critical challenge and the need

of highly advanced computational methods that can

automate, improve, and expedite the process of

pancreatic cancer detection. Figure 1 shows the

pancreas.

In this paper, we provide a systematic

methodology for the diagnosis of pancreatic cancer

using CNN in MATLAB. MATLABs Deep Studying

Toolbox and Picture Processing Toolbox present a

built-in platform for knowledge preprocessing,

mannequin coaching, and analysis, which is extra

accessible for individuals in academia and the trade.

It explores different CNN architectures for

classification and segmentation tasks, and how

methods such as multimodal data fusion and transfer

learning can improve performance in detection. By

leveraging in doing this by ways of research, we

would be able to contribute towards the building of

automated and reliable diagnostics tools for

pancreatic cancer which would in turn help in earlier

diagnosis and thus combating this disease in a much

better way.

Figure 1: Pancreas.

2 LITERATURE SURVEY

Recent advancements in deep learning have opened

new avenues for the early detection and accurate

Akhila, S., Swamy, G. N., Sahithi, S., Faheem, S. and Rajesh, J.

Pancreatic Cancer Detection Using Deep Learning.

DOI: 10.5220/0013932300004919

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 5, pages

519-524

ISBN: 978-989-758-777-1

519

diagnosis of pancreatic cancer, one of the deadliest

forms of cancer due to its asymptomatic nature in early

stages. Chen, Z, Wang, L & Huang Y: developed a

CNN-based model focused on early detection of

pancreatic cancer using CT images. The study

emphasizes the use of image preprocessing to

improve the visibility of pancreatic tumors, which are

often difficult to distinguish in standard CT scans Liu,

X., Zhao, J., &Li, F: explored the segmentation of

pancreatic lesions using a deep learning-based UNet

model. This study applied the UNet architecture to

isolate and segment tumors from MRI images,

achieving improved accuracy in identifying lesion

boundaries Wang, & Chen H Zhu L.: proposed a

classification method for pancreatic cancer using

transfer learning with VGG16. The study utilized a

small dataset and leveraged pretrained VGG16 layers,

which reduced training time and computational

requirements Ibrahim, H., Khan, M., & Ali, Z:

focused on using MATLAB’s Deep Learning

Toolbox to detect gastrointestinal cancers, including

pancreatic tumors. Zhou, Xie & Wang, G:

implemented a deep learning framework to

automatically detect pancreatic cancer from

endoscopic ultrasound images. Zhu, J Gao M & Sun

R: introduced a multi-modal deep learning approach

combining MRI and CT scans to improve pancreatic

cancer detection accuracy. By fusing features from

both imaging modalities, the model provided more

reliable diagnostic outcomes, highlighting how multi-

modal data enhances deep learning model

performance. Singh, R., Kumar, the surveyed

literature highlights the potential of deep learning as

a transformative tool for the detection of pancreatic

cancer.Yang, W., Zhang, H This study presents a

CNN model that analyzes multi- phase CT scans for

pancreatic cancer detection. The authors highlight

how phase-specific feature extraction improves

classification accuracy. Ma, J., He, Z The research

proposes a hybrid deep learning model that integrates

CNNs and recurrent neural networks (RNNs) to

capture both spatial and sequential features in medical

imaging.

Shen, C., Yu, Y., & Wang The study explores the

application of ResNet-based transfer learning in

pancreatic cancer classification. The authors

demonstrate that pretrained ResNet models can

effectively classify pancreatic tumors in MRI scans,

achieving high sensitivity and specificity with

minimal training data. Gupta, P., Singh, R., & Kaur

the proposed model processes entire 3D scans rather

than 2D slices, improving tumor identification and

reducing segmentation errors. Kwon, S., Lee, J., &

Park The research introduces a self-supervised

learning technique that pre- trains deep learning

models using unlabeled medical images before fine-

tuning with labeled data. Zhang, T., Li, M., & Chen

This study focuses on improving the interpretability

of deep learning models used for pancreatic cancer

detection. The authors integrate explainable AI

techniques such as Grad-CAM to highlight critical

tumor regions in CT and MRI scans, enhancing

clinician trust in AI-driven diagnosis. Patel, N.,

Ghosh the study applies a U-Net model for automated

pancreatic tumor segmentation in endoscopic

ultrasound images. The authors refine the

segmentation process by incorporating adaptive

thresholding techniques, leading to improved tumor

boundary delineation. Wu, H., Fan This research

explores multi-modal deeplearning approaches,

combining radiomic features from CT, MRI, and PET

scans. The study demonstrates that fusing multiple

imaging modalities enhances classification accuracy

and improves early-stage detection. CNNs have

shown strong capabilities in identifying cancerous

patterns in imaging data, and when combined with

techniques such as transfer learning and multimodal

analysis, these models can achieve high levels of

diagnostic accuracy. UNet- based segmentation has

proven essential for accurately isolating tumors

within the pancreas, aiding in both diagnosis and

treatment planning. The use of MATLAB has enabled

researchers to efficiently implement and experiment

with these models, streamlining the workflow from

data preprocessing to model evaluation.

3 PROPOSED SYSTEM

The system presented herein employs a

Convolutional Neural Network (CNN) to facilitate

the automatic detection of pancreatic cancer from

medical images, primarily CT and MRI scans. The

primary objective is to augment early diagnosis

precision through the analysis of these images to

distinguish between cancerous and non- cancerous

pancreatic tissue. The developmental framework is

MATLAB, integrating the Deep Learning Toolbox

and Image Processing Toolbox to ensure an

optimized and efficient methodology.

3.1 Data Collection and Preparation

A reliable dataset of pancreatic images is obtained

from the existing medical sources or databases to

validate the performance of the system. This dataset

is carefully annotated to distinguish between healthy

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

520

and cancer tissues. Preprocessing is an essential

stage that consists of:

 Resizing: This is a uniform scaling of images so

that it will meet the input requirements of the

CNN model.

 Normalization: Scaling pixel values to a

standard range (usually 0-1) to help the model

converge faster during training.

 Regularization: Tumorous regions in the images

are enhanced using enhancement techniques to

improve their visibility.

 Data Augmentation: A technique to reduce

overfitting that is especially useful in low-data

scenarios by splitting up the dataset into smaller

subsets by applying rotations, horizontal

flipping, cropping, and other transformative

functions.

3.2 CNN Model Architectural Design

The CNN architecture is meticulously crafted to

classify images into cancerous and non-cancerous

categories by identifying and learning pertinent

features of pancreatic cancer. It comprises the

following layers: Input Layer: This layer accepts

the preprocessed images for analysis.

 Convolutional Layers: Multiple layers equipped

with varying kernel sizes are employed to extract

both low- level (e.g., edges) and high-level

features (e.g., shapes and textures).

 Activation Function: ReLU activation is

sequentially applied post-convolution to

introduce non-linearity and enhance the model's

pattern recognition capabilities.

 Pooling Layers: These layers serve to reduce the

spatial dimensions of the feature maps, thereby

capturing essential features and minimizing

computational complexity.

 Fully Connected Layers: These layers are

responsible for connecting the features extracted

from the convolutional layers to the final output.

Dropout layers are interspersed to prevent

overfitting.

 Output Layer: Depending on the classification

task (binary or multi-class), a softmax or

sigmoid function is utilized to generate the final

predictions.

3.3 Transfer Learning for Model

Training

Given the scarcity of pancreatic cancer-specific data,

transfer learning is proposed to capitalize on the

knowledge of pretrained CNN models (e.g., VGG16,

ResNet). The base model's layers are fine-tuned with

the pancreatic dataset to optimize accuracy and

reduce training time.

3.4 Model Evaluation and Validation

To assess the system's performance, several metrics

are computed, including accuracy, sensitivity,

specificity, and F1 score. K-fold cross-validation is

performed to evaluate the model's robustness across

various dataset partitions and minimize overfitting.

3.5 System Deployment and MATLAB

Interface

A user- friendly MATLAB GUI is developed to

facilitate system deployment. Users, such as

radiologists and clinicians, can upload images for

analysis, which are then processed by the trained CNN

model. The system returns classification results with

a probability of cancer presence, enhancing

accessibility and clinical integration.

3.6 Potential Future Enhancements

Multimodal Imaging Integration: The system may be

further developed to incorporate multiple imaging

modalities, such as combining MRI and CT data, to

improve diagnostic accuracy.

Segmentation for Tumor Localization:

Incorporating segmentation models, like UNet, can

refine the system's ability to pinpoint the exact

location of the tumor within the pancreas, thereby

enriching the interpretability of the findings. Figure 2

block diagram.

Figure 2: Block Diagram.

3.7 Algorithms

Step 1: Choose a right framework and install it. //tensor

flow addons is taken as a framework.

Pancreatic Cancer Detection Using Deep Learning

521

Step 2: Read the CSV file as input data.

Step 3: choose parameters from the taken dataset

Drop few columns like sample id, patient cohort,

sample origin, stage, benign sample diagnosis.

//Features required to diagnose are selected. Replace

the values: If Gender = ‘M’: Set as 1 If Gender = ‘F’:

Set as 0

Step 4: The data will be partitioned into training sets

and test sets.

Step 5: Creating model. In the first dense layer apply

RELU gradient on the data f(x) =1/(1-e^x)

Step 6: In the second dense layer activation on the data.

Step 7: Test the trained model using testing set.

Step 8: compare the new model with any existing

model Check accuracy, precision, recall, f1 score from

the graphs.

4 SYSTEM ARCHITECTURE

4.1 Training Process

Figure 3: Training Flow Chart.

4.2 Testing Process

Figure 4: Testing Flow Chart.

Figure 3 and 4 shows the training and evaluation

framework for the proposed pancreatic cancer

detection system is predicated on the utilization of a

Convolutional Neural Network (CNN) model, which

is executed within the MATLAB environment. This

model leverages transfer learning from a pre-existing,

high- performance architecture such as VGG16 or

ResNet. During the training phase, the system

processes Computed Tomography (CT) or Magnetic

Resonance Imaging (MRI) images, which have been

meticulously labeled and annotated for the presence

of pancreatic cancer. These images undergo

preprocessing and data augmentation to generate a

robust and comprehensive dataset capable of

capturing a broad spectrum of variations inherent in

medical imaging. The CNN model is then

systematically trained on this dataset, with specific

layers optimized to discern unique patterns indicative

of pancreatic malignancies.

5 RESULTS

Figure 5: Model Training Results.

Figure 6: Training Accuracy Progression.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

522

Figure 7: Training Loss Progression.

Figure 5 shows the model training results 7 and 6

shows the training accuracy and loss progression.

Figure 8 and 9 shows the tumor detected and no

tumor.

Figure 8: Tumor Detected.

Figure 9: No Tumor.

6 CONCLUSIONS

This plot represents the below model which has

shown high accuracy in pancreatic tumors

classification: A simple under fitting AlexNet based

CNN model achieved accuracy of 100% on test

dataset Such high accuracy implies that normal

pancreatic tissue and pancreatic tumors may be

identified with a high confidence level by the deep

learning model, providing a powerful diagnostic

assistant in the clinical setting. This system could help

radiologists in classifying CT scans and could play an

important role in the early detection and accurate

diagnosis of pancreatic cancer, addressing a critical

need in oncology. In addition, the system's reliable

performance is promising for real deployment in

clinical settings to ease the burden of medical staff

while providing timely and precise diagnosis for

patients. Future work may investigate the use of

similar deep learning approaches with other types of

medical imaging data or the development of models

that are truly robust to a broader range of cases and

conditions to further increase its generalizability and

clinical utility.

7 FUTURE SCOPE

In conclusion, the future of detecting pancreatic

cancer through deep learning in MATLAB looks

promising with immense potential for growth and

advancements. Domain research may focus on the

integration of multimodal imaging data, including

MRI, CT and PET scans to utilize a comprehensive

set of features that could improve diagnostic

accuracy. New CNN formulations, everywhere from

new small-scale AI architectures like 3D CNNs or an

amalgam of CNNs with other deep learning

approaches will be able to address more complex

tumor detection and localization. Additionally, scale

the dataset diversity especially annotated image from

wider demography population will leads to model

generalization. The integration of interpretability

tools to explain model predictions to medical

practitioners is thus of utmost importance to allow

for clinical adoption. Ultimately, this approach can

take advantage of the compatibility of the MATLAB

environment with clinical systems and encourage its

in-time use, and thus assist radiologists in an early

detection of pancreatic malignancies and,

consequently, improved patient outcomes: timely and

accurate diagnostics.

REFERENCES

Chen, Z., Wang, L., & Huang, Y. (2021). CNN- based

approach for early detection of pancreatic cancer in CT

scans.

Gupta, P., Singh, R., & Kaur, A. (2021). Automated

Pancreatic Tumor Detection Using Deep Learning and

Pancreatic Cancer Detection Using Deep Learning

523

3D CNNs. Artificial Intelligence in Medicine, 121,

102200.

Ibrahim, H., Khan, M., & Ali, Z. (2022). MATLAB-based

deep learning approach for detecting gastrointestinal

cancers.

Kwon, S., Lee, J., & Park, H. (2020). Self- Supervised

Learning for Pancreatic Cancer Detection in CT

Images. Medical Image Analysis, 64, 101731.

Liu, X., Zhao, J., & Li, F. (2020). Segmentation of

pancreatic lesions using a deep learning-based UNet

model.

Ma, J., He, Z., & Xu, F. (2020). A Hybrid Deep Learning

Framework for Pancreatic Tumor Classification. IEEE

Transactions on Biomedical Engineering, 67(9), 2563–

2574.

Patel, N., Ghosh, S., & Das, R. (2019). Deep Learning-

Based Segmentation and Classification of Pancreatic

Tumors in Endoscopic Ultrasound Imaging.

Biomedical Signal Processing and Control, 52, 123–

130.

Shen, C., Yu, Y., & Wang, X. (2022). Transfer Learning

with ResNet for Pancreatic Cancer Detection in MRI.

Computers in Biology and Medicine, 134, 104532.

Singh, R., Kumar, A., & Gupta, S. (2019). Automated

pancreatic tumor segmentation using deep learning and

MATLAB implementation.

Wang, Y., Chen, H., & Zhu, L. (2019). Transfer learning

for pancreatic cancer classification using VGG16.

Wu, H., Fan, Y., & Zhao, Q. (2022). Multi-Modal Fusion

for Pancreatic Cancer Detection Using Deep Learning.

Journal of Imaging Science and Technology, 66(2),

034502.

Yang, W., Zhang, H., & Liu, Q. (2021). Deep Learning-

Based Pancreatic Cancer Diagnosis from Multi-Phase

CT Scans. Journal of Medical Imaging and Health

Informatics, 11(4), 1234–1241.

Zhang, T., Li, M., & Chen, W. (2021). Explainable AI for

Pancreatic Cancer Diagnosis Using Deep Learning.

Nature Scientific Reports, 11, 12893.

Zhou, Y., Xie, L., & Wang, G. (2020). Deep learning-based

system for automatic detection of pancreatic cancer in

endoscopic ultrasound images.

Zhu, J., Gao, M., & Sun, R. (2021). Multi-modal deep

learning for pancreatic cancer detection using MRI and

CT fusion.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

524