МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
397
ADVANCED AI-DRIVEN FRAMEWORK FOR LUNG CANCER DETECTION
USING CT SCAN IMAGES AND MOBILE APPLICATION INTEGRATION
1
Xusanov K.X.,
2
Nematov N.F.
Muhammad al-Xorazmiy nomidagi Toshkent axborot texnologiyalari universiteti
Samarqand filiali
ABSTRACT
Lung cancer is among the leading causes of cancer-related deaths globally,
primarily due to late-stage detection. Recent advancements in Artificial Intelligence (AI)
and deep learning have opened new pathways for early and accurate diagnosis of lung
cancer, addressing the limitations of traditional diagnostic methods. This study
introduces an enhanced approach for processing lung CT (Computed Tomography) scan
images using deep convolutional neural networks (CNNs) integrated with advanced
noise reduction and feature extraction techniques. Unlike prior studies, our model
dynamically applies adaptive thresholding using k-means clustering and morphological
operations to isolate the lung region effectively and identify cancerous nodules.
Additionally, we integrate 3D image meshing for improved visualization and analysis.
To bridge the gap between diagnosis and patient awareness, the proposed model is
supported by a mobile application ecosystem, enabling real-time health monitoring and
communication. This research further emphasizes the potential of AI-based
methodologies in detecting lung cancer at earlier stages, thereby reducing mortality
rates. The paper also highlights directions for future advancements, such as staging
cancerous tissues and improving diagnostic precision through more robust training
datasets.
Keywords:
Lung Cancer, Artificial Intelligence, Deep Learning, CNN, CT Scan,
Mobile Health Applications.
1
. INTRODUCTION:
Lung cancer remains one of the most challenging and fatal
diseases worldwide, accounting for a significant proportion of cancer-related deaths
annually. Its insidious progression and delayed onset of symptoms often result in
diagnosis at advanced stages, where treatment options are limited, and survival rates
drastically decrease. Lung cancer is primarily categorized into two types: small cell lung
cancer (SCLC) and non-small cell lung cancer (NSCLC), with NSCLC being the most
prevalent, constituting 80–85% of cases. Early detection of cancerous nodules within
lung tissues is critical for effective treatment and improved patient outcomes.
The advent of Artificial Intelligence (AI) and advancements in digital imaging
techniques have revolutionized medical diagnostics, offering new tools for more
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
398
accurate and efficient detection of diseases. Traditional diagnostic approaches in
medical imaging often rely on manual interpretation and handcrafted features, which
can be prone to variability and error. Recent developments in deep learning, particularly
convolutional neural networks (CNNs), provide a robust alternative by enabling
automated feature extraction, classification, and prediction with high precision.
This research builds on prior efforts by proposing a comprehensive AI-powered
system for lung cancer detection using CT scan images. Our approach incorporates
advanced preprocessing techniques, including noise removal, adaptive segmentation,
and dynamic thresholding, to enhance the quality of medical images for analysis. The
integration of CNNs facilitates the detection of cancerous nodules, while 3D meshing
and visualization improve diagnostic clarity.
In addition to imaging advancements, this study also emphasizes patient-centric
care through mobile applications. These applications not only serve as tools for
diagnosis and monitoring but also play a pivotal role in patient education,
communication, and mental health support. By bridging the gap between AI-driven
diagnostics and patient interaction, we aim to create a holistic solution for managing
lung cancer.
The remainder of this paper discusses the related work, the proposed
methodology, experimental results, and potential areas for future research. We highlight
the significance of AI-based innovations in reducing diagnostic delays and mortality
rates associated with lung cancer.
2. RELATED WORK:
Lung cancer detection has been a focal area of research
in medical imaging due to its high mortality rate and the complexities associated with
its early diagnosis. Traditional methods rely on handcrafted features and rule-based
systems, which often lack the precision and scalability required for large datasets. With
advancements in artificial intelligence (AI), researchers have shifted towards automated
systems, leveraging machine learning (ML) and deep learning (DL) techniques to
improve detection accuracy and efficiency.
Several studies have demonstrated the potential of AI in medical imaging. For
instance, early applications of image processing techniques utilized pixel-based filtering
methods, such as Gaussian and median filters, to enhance image quality and reduce
noise. These approaches were combined with segmentation algorithms, such as
watershed and thresholding methods, to isolate regions of interest (ROIs) in CT images.
While effective in reducing visual clutter, these methods were limited in detecting small
or irregularly shaped nodules, which are critical for early cancer diagnosis.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
399
Recent research has focused on integrating deep learning techniques, particularly
convolutional neural networks (CNNs), for lung nodule detection and classification.
CNN-based models have shown significant promise due to their ability to learn
hierarchical features directly from raw image data. For example, models trained on
large-scale datasets have achieved notable improvements in sensitivity and specificity
compared to conventional machine learning classifiers like support vector machines
(SVM) and random forests. Despite these advancements, achieving 100% accuracy
remains a challenge, as many models struggle with false positives and false negatives.
Studies such as those by Makajua et al. (2018) explored the application of CNNs
for CT image segmentation and classification. Their research introduced median and
Gaussian filtering techniques in the preprocessing stage, coupled with a Keras-based
neural network for classification. However, the limitations in segmentation quality and
computational efficiency highlighted the need for more robust and scalable models.
Similarly, works by Ait Skourt et al. (2018) utilized advanced neural network
architectures for lung nodule segmentation, achieving improved accuracy but still facing
challenges in real-time applications due to computational overhead.
Another notable approach involved K-means clustering for color image
enhancement and cell segmentation. Although this method provided insights into
segmentation improvements, its application to complex lung structures required further
refinement. Furthermore, studies incorporating morphological operations and noise
removal techniques demonstrated significant improvements in image preprocessing but
highlighted the need for dynamic and adaptive methodologies for handling diverse
datasets.
This study addresses these gaps by proposing a novel framework that combines
adaptive noise removal, dynamic thresholding, and deep convolutional neural networks
for lung cancer detection. In addition, the use of 3D image meshing and advanced
visualization techniques provides enhanced clarity for medical practitioners. By
integrating these technical advancements with mobile applications for patient
interaction, this research aims to bridge the existing gaps in lung cancer diagnosis and
treatment.
3. PROPOSED MODEL:
The proposed model aims to enhance the accuracy and
efficiency of lung cancer detection through a comprehensive pipeline combining
advanced image preprocessing, segmentation, and deep learning techniques. This
section outlines the key components of the model, from data acquisition to feature
extraction and analysis.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
400
3.1 Data acquisition:
The model utilizes CT scan images in the DICOM (Digital
Imaging and Communications in Medicine) format. These images contain detailed
volumetric data that allow for a thorough analysis of lung structures. The dataset used
for this study is sourced from the Lung Image Database Consortium (LIDC-IDRI),
which includes annotated cases of lung nodules and other thoracic structures.
3.2 Image preprocessing:
Effective preprocessing is vital for accurate results in
medical image analysis. The following steps are performed:
•
Noise removal
: Gaussian and median filters are applied to eliminate
common noise types such as Gaussian noise, salt-pepper noise, and speckle noise.
•
Dynamic thresholding
: Adaptive threshold values are calculated using the
k-means clustering algorithm (k=2). This allows for isolating relevant features such as
lung tissues while excluding non-relevant regions like bones and air pockets.
•
Morphological operations
: Dilation and erosion techniques are used to
refine the lung region by filling gaps and removing unwanted artifacts.
3.3 Image segmentation:
The segmented images are processed to focus solely on
the region of interest (ROI)—the lung area. 3D slicing at the voxel level is employed for
better visibility and analysis. This step ensures that only the relevant portions of the CT
scans are analyzed, reducing computational load and improving detection precision.
3.4 Deep learning application:
A deep convolutional neural network (CNN) is
employed for feature extraction and classification. The proposed architecture includes:
•
Input layer
: Grayscale images of size 512×512 are fed into the network.
•
Convolutional layers
: These layers extract spatial features using learnable
filters.
•
Pooling layers
: Max pooling is used to reduce spatial dimensions,
preserving essential features while minimizing computational costs.
•
Fully connected layers
: These layers analyze the extracted features to
predict the presence of cancerous nodules.
•
Softmax output
: Probabilistic values are computed to classify the image into
categories such as benign or malignant nodules.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
401
Figure 1. Proposed model for ct scan image processing
3.5 3D Image visualization:
To aid medical practitioners in diagnosis, the model
generates 3D reconstructions of the lung cavity. Image meshing techniques are applied
to the segmented slices, providing an intuitive visual representation of the lung structure.
Figure 2. 3D Meshed representation of the lung cavity
3.6 Mobile application integration: To
bridge the gap between diagnosis and
patient care, the model incorporates a mobile application interface. This interface
enables patients to monitor their health status, access educational resources, and
communicate seamlessly with healthcare providers.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
402
Figure 3. Mobile application workflow for patient interaction
3.7 Key advantages:
The model dynamically adjusts to noise levels, ensuring
precise segmentation and feature extraction.
•
It handles large datasets efficiently, making it suitable for real-time clinical
applications.
•
3D visualization enhances diagnostic clarity, facilitating better decision-
making by healthcare professionals.
4. ARCHITECTURE
The overall architecture of the proposed model integrates preprocessing, deep
learning, and visualization to deliver a robust lung cancer detection system. The process
flow is divided into distinct phases:
1.
Data input
: DICOM CT scan images are loaded, converted to grayscale,
and preprocessed for analysis.
2.
Preprocessing and segmentation
: Noise removal and morphological
operations refine the image quality. Adaptive segmentation isolates the lung region for
further study.
3.
Feature extraction and classification:
A convolutional neural network
(CNN) extracts features, classifies nodules as benign or malignant, and provides
diagnostic insights.
4.
Visualization and Interaction: Processed images are rendered in 2D and 3D
formats, while mobile applications support patient communication and education.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
403
Figure 4. Overall system architecture
5. EXPERIMENTAL RESULTS
5.1 Dataset and experimental setup:
The experiments are conducted using 130
CT scan images from the LIDC-IDRI database. The DICOM images are converted into
2D grayscale slices and processed using Python libraries such as NumPy, SciKit-Image,
and TensorFlow. The training dataset is split into 70% training and 30% testing subsets.
5.2 Evaluation metrics
The performance of the model is evaluated using the following metrics:
•
Accuracy
: Proportion of correctly identified nodules.
•
Precision
: Ability to identify only relevant cancerous nodules.
•
Recall (Sensiti
vity): Ability to identify all cancerous nodules.
•
F1-Score
: Harmonic mean of precision and recall.
5.3 Results and analysis
The model demonstrates the following performance metrics:
•
Accuracy: 96.5%
•
Precision: 94.8%
•
Recall: 92.7%
•
F1-Score: 93.7%
These results indicate that the proposed model effectively detects lung cancer
nodules with high precision and minimal false positives.
5.4 Comparative analysis
The proposed model outperforms traditional methods and earlier AI-based models
in terms of accuracy and computational efficiency. For example:
•
A previous approach using Random Forest classifiers achieved 59.2% sensitivity
and 66% efficiency.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
404
•
The proposed CNN-based model significantly improves these metrics,
particularly in reducing noise-related false positives.
6. CONCLUSION AND FUTURE WORK
The study introduces a comprehensive framework for lung cancer detection
leveraging AI and mobile applications. The integration of advanced image processing
techniques and deep learning enhances diagnostic accuracy while reducing false
positives. The use of 3D visualization further improves the clarity of results, aiding
medical professionals in effective decision-making.
Future work includes:
1.
Expanding the dataset to improve model robustness across diverse patient
demographics.
2.
Enhancing the CNN architecture to detect cancer stages and early signs
with greater precision.
3.
Incorporating federated learning approaches for secure and distributed
training on hospital datasets.
4.
Developing more interactive and intelligent mobile applications for patient
care and real-time monitoring.
This research emphasizes the transformative potential of AI in medical
diagnostics and highlights the importance of patient-centric solutions for improving
healthcare outcomes.
REFERENCES
1.
American Cancer Society. "What Is Lung Cancer?" [Online]. Available:
https://www.cancer.org/cancer/lung-cancer/about/what-is.html. Accessed: Dec. 2024.
2.
Geert Litjens, Thijs Kooi, Babak Ehteshami Bejnordi, et al. "A Survey on Deep
Learning in Medical Image Analysis,"
Medical Image Analysis
, vol. 42, pp. 60–88,
2017. DOI: 10.1016/j.media.2017.07.005.
3.
Suren Makajua, et al. "Lung Cancer Detection Using CT Scan Images,"
Indian
Journal of Science and Technology
, vol. 11, no. 40, pp. 1–9, 2018. DOI:
10.17485/ijst/2018/v11i40/120482.
4.
Brahim Ait Skourt, Abdelhamid El Hassani, Aicha Majda. "Lung CT Image
Segmentation Using Deep Neural Networks,"
Procedia Computer Science
, vol. 127, pp.
109–117, 2018. DOI: 10.1016/j.procs.2018.01.104.
5.
Kamelia Roy, et al. "A Comparative Study of Lung Cancer Detection Using
Supervised Neural Network,"
International Journal of Signal and Image Processing
,
vol. 1, no. 1, pp. 1–10, 2017. DOI: 10.21742/ijsesv.2017.1.1.01.
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
405
6.
Man Yan, Jianyong Cai, et al. "K-means Cluster Algorithm Based on Color Image
Enhancement for Cell Segmentation,"
Journal of Biomedical Imaging
, vol. 12, pp. 45–
52, 2019.
7.
Weixing Wang, Shuguang Wu. "A Study on Lung Cancer Detection by Image
Processing,"
Journal of Engineering Science and Technology Review
, vol. 10, no. 3, pp.
62–68, 2018. DOI: 10.25103/jestr.103.10.
8.
S. Kalaivani, et al. "Lung Cancer Detection Using Digital Image Processing and
Artificial Neural Networks,"
Journal of Global Engineering
, vol. 5, no. 2, pp. 35–40,
2019. DOI: 10.13052/jge1904-4720.521.
9.
Selin Uzelaltinbulata, Buse Ugur. "Lung Tumor Segmentation Algorithm,"
9th
International Conference on Theory and Application of Soft Computing
, Budapest,
Hungary, 2017. DOI: 10.1016/j.procs.2017.11.010.
10.
Bhawna Goyal, Ayush Dogra, Sunil Agrawal, B.S. Sohi. "Noise Issues Prevailing
in Various Types of Medical Images,"
Panjab University Journal of Imaging Science
,
vol. 20, pp. 78–85, 2018.
11.
Armato SG III, et al. "The Lung Image Database Consortium (LIDC) and Image
Database Resource Initiative (IDRI): A Completed Reference Database of Lung
Nodules on CT Scans,"
Medical Physics
, vol. 38, no. 2, pp. 915–931, 2011. DOI:
Clark K, et al. "The Cancer Imaging Archive (TCIA): Maintaining and Operating
a Public Information Repository,"
Journal of Digital Imaging
, vol. 26, no. 6, pp. 1045–
1057, 2013. DOI: 10.1007/s10278-013-9622-7.
13.
К Хусанов, М Ахроров, Ж Тошбоев “ Kompyuter arxitekturasi” fanidan mobil
ilova axborot tizimini ishlab chiqish” Информатика и инженерные технологии vol.
1, issue 2, pp 13-17, 2023/11/7
14.
Hoshimova Nilufar, Husanov Kamoliddin “SUN’IY INTELLEKTNING
TIBBIYOTDA QO ‘LLANILISHI VA AFZALLIKLARI” International conference on
multidisciplinary science vol. 1, issue 5, pp 78-81, 2023/11/22
МЕДИЦИНА, ПЕДАГОГИКА И ТЕХНОЛОГИЯ:
ТЕОРИЯ И ПРАКТИКА
Researchbib Impact factor: 11.79/2023
SJIF 2024 = 5.444
Том 2, Выпуск 11, Ноябрь
https://universalpublishings.com
406
