Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725

Computer-Assisted Lung Cancer Diagnosis

through Morphological Analysis & CNN
I. Saleth Mary1; Dr. A. Shanthasheela2
1
Research Scholar, Department of Computer Science, Mother Teresa Women’s University, Kodaikanal
Assistant Professor, Department of Computer Science, St. Antony’s College of Arts and Sciences for
Women, Thamaraipadi, Dindigul, Tamil Nadu, India.
2
Assistant Professor, Department of Computer Science, M.V.Muthiah Government Arts College for Women,
Dindigul – 624001, India.

Publication Date: 2025/05/10

Abstract: Lung cancer is an unregulated development of cells that begins in the lung and spreads to other parts of the body,
posing a significant risk to human life. Radiological imaging, such as computed tomography (CT) scans and X-rays, is the
primary tool for diagnosing lung cancer. However, a person's ability to interpret a large number of CT images might vary
greatly, especially when the scans show many gray level fluctuations. The purpose of this study is to use Python-based
machine learning and image processing approaches to detect lung cancer. Using the National Center for Cancer Diseases
lung cancer dataset, this paper analyzes lung scans to determine if they are malignant or non-cancerous. Based on the study's
top-performing solution, the code first preprocesses the images before applying segmentation and feature extraction
techniques. The suggested approach makes a cancer prediction based on retrieved properties that were obtained through
morphological processing.

Keywords: Lung Cancer, Segmentation, Malignant, Morphological Operations, CNN.

How to Cite: I. Saleth Mary; Dr. A. Shanthasheela. (2025). Computer-Assisted Lung Cancer Diagnosis through Morphological
Analysis & CNN. International Journal of Innovative Science and Research Technology, 10(4),
3127-3133. https://doi.org/10.38124/ijisrt/25apr1725.

I. INTRODUCTION interpretation. To address this, automated diagnostic systems

have been developed using machine learning and image
Cellular breakdown in the lungs remains as the chief processing techniques, aiming to enhance both the accuracy
reason for disease related mortality around the world, and efficiency of lung cancer detection [2].
comprising a critical worldwide general wellbeing concern
and positioning as the essential malignant growth in men and Among these technologies, Convolutional Neural
the second most predominant in females. Unchecked cell Networks (CNNs) stand out as a powerful type of deep
development in the lungs is the hallmark of lung cancer, a learning algorithm, particularly well-suited for image
potentially fatal illness. Its late discovery and diagnosis recognition and processing tasks. A CNN typically consists
contribute significantly to its status as one of the world's top of several key layers, including convolutional layers, pooling
causes of cancer-related fatalities [1]. Although early layers, and fully connected layers. Its architecture is inspired
detection greatly increases the chances of a successful course by the visual processing mechanisms of the human brain,
of therapy and survival, it can be difficult due to the making it highly effective at identifying spatial relationships
complexity of lung tissues and the subtlety of early-stage and hierarchical patterns within images [3][4].
cancer indications. Computed tomography (CT) scans and
other radiological imaging methods are essential for the The core components of a Convolutional Neural
diagnosis of lung cancer. Despite its importance, radiologists Network include:
face significant challenges in lung cancer diagnosis due to the
vast amount of data involved and the need for precise

IJISRT25APR1725 www.ijisrt.com 3127

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725

Fig 1: The Architecture of Convolutional Neural Networks

 Convolutional Layers: These layers apply filters, also These methods are very helpful in differentiating
known as kernels, to the input images through between non-cancerous and malignant tissues based on the
convolutional operations, enabling the detection of size, texture, and form of nodules. Morphological approaches
features such as edges, textures, and more complex have the advantage of enhancing the traits of interest and
patterns. Convolution helps preserve the spatial suppressing extraneous details, which results in more
relationships between pixels, which is essential for accurate and dependable cancer detection. Automated lung
accurate image analysis. cancer detection systems can achieve far better diagnosis
 Pooling Layers: Pooling layers reduce the spatial results by combining morphological processing with other
dimensions of the feature maps, decreasing the image processing and machine learning methods.
computational load and the number of parameters in the
network. One commonly used method is max pooling, There are numerous Morphological Analysis [6][7][8].
which selects the highest value from a set of neighboring Some of them are given below:
pixels, effectively summarizing the most prominent
features.  Facilitates creative problem-solving by offering multiple
 Activation Functions: Non-linear activation functions, solutions.
such as the Rectified Linear Unit (ReLU), introduce non-  Encourages collaboration and improves communication
linearity into the model. This enables the network to learn by involving team members.
complex and non-linear patterns in the data, which is  Aids in strategic business analysis.
critical for tasks like image classification.  Enhances search engine results by identifying un-lexicon
 Fully Connected Layers: These layers interpret the high- words.
level features extracted by earlier layers and are  Enhances biological insights by reducing labeling reagent
responsible for making final predictions. In fully artifacts.
connected layers, each neuron is connected to every  Provides unbiased analysis of cell morphology.
neuron in the previous layer, allowing for comprehensive  Produces repeatable experimental results.
integration of the learned features.
II. RELATED WORK
This paper proposes a comprehensive approach for
detecting bright regions in lung images—often indicative of Recent advancements in deep learning have
lesions—by integrating techniques such as noise reduction, significantly improved the accuracy of medical image
contrast enhancement, segmentation, and classification, analysis. Convolutional Neural Networks (CNNs) have been
powered by advances in machine learning and image widely used for lung cancer detection due to their ability to
processing. learn hierarchical features from images. U-Net, Mask R-
CNN, and 3D CNNs have shown promising results in
Morphological analysis plays a vital role in lung cancer segmenting and classifying lung nodules. Transfer learning,
detection, as it focuses on the shape and structure of using pre-trained models such as ResNet, VGG, and Efficient
biological tissues in medical imaging. Techniques such as Net, has also been applied to medical imaging tasks, reducing
dilation, erosion, and edge detection in CT imaging help the need for large annotated datasets.
highlight and isolate regions of interest, thereby aiding in the
accurate identification of potential cancerous areas [5].

IJISRT25APR1725 www.ijisrt.com 3128

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725
Despite these advancements, challenges remain in  Lung Segmentation
differentiating between benign and malignant nodules, Lung regions are segmented using Sobel edge detection
especially in datasets with imbalanced class distributions. combined with morphological operations. The Sobel operator
Traditional machine learning methods, such as Support detects the lung boundaries, while morphological dilation
Vector Machines (SVM) and Random Forests, have been helps to close gaps and fine-tune the edges. This ensures
used for lung cancer detection but often lack the robustness precise isolation of the lung area, which is critical for reliable
of deep learning approaches. This paper builds on these works analysis.
by proposing a hybrid approach that combines morphological
analysis with CNNs to improve the accuracy and  Feature Extraction
generalizability of lung cancer detection.
 Morphological features—such as the size, shape, and
III. PROPOSED METHODOLOGY texture of the segmented regions—are extracted to
characterize nodules. These features play a crucial role in
In order to precisely identify malignant spots in CT distinguishing between benign and malignant tissue.
image, the suggested method for lung cancer detection
combines CNN with image processing approaches. First, the  Classification
images are preprocessed using median filtering to minimize The classification is performed using a Convolutional
noise. Following initial image processing, thresholding and Neural Network (CNN) composed of five convolutional
morphological operations are applied to segment the lung layers, followed by max-pooling layers and fully connected
regions and isolate any nodules present in the CT scans [9]. layers. The network is trained to recognize complex patterns
Morphological dilation is then employed to further refine the in the extracted features and accurately classify the lung
boundaries by closing gaps and emphasizing regions of tissue.
interest. Once segmentation is complete, feature extraction is
performed to gather critical information about the size, shape, The Rectified Linear Unit (ReLU) activation function is
and distribution of the detected nodules. These morphological used to introduce non-linearity, and dropout is applied to
features serve as input for the classification stage, where the prevent over fitting. Transfer learning is employed using the
system predicts whether the lung tissue is malignant or ResNet-50 model, which was pre-trained on the Image Net
benign. dataset. The model is fine-tuned using the lung cancer dataset,
with hyper parameters optimized through grid search.
A. Proposed Methodology
The proposed approach involves several key stages:  Algorithm
preprocessing, lung segmentation, feature extraction, and Here's a step-by-step breakdown of the algorithm to
classification. Each phase plays a vital role in ensuring detect the uploaded image is malignant or normal.
accurate detection and diagnosis of lung abnormalities in CT
images.  Load Image
 Load pre-trained model
 Dataset  Perform prediction by the following steps
This study utilizes the dataset provided by the National  Preprocess the image with filtering for noise reduction
Center for Cancer Diseases, comprising 1,000 CT scan initially and then perform thresholding followed by
images—600 from malignant cases and 400 from benign morphological operation
cases. The scans were sourced from various hospitals to  Detect Lung Sides such as left and right side of lungs
ensure diversity in scanner types, image resolutions, and  Segment lung image and extract left and right lungs
patient demographics. To address class imbalance and reduce
 Extract features from a sub-region of the lung
the risk of over fitting, data augmentation techniques such as
rotation, flipping, and scaling were employed.  Check feature and vector for prediction.
 Display Normal lung or Tumor present based on feature.
 Preprocessing
Preprocessing begins with noise reduction and contrast  Block Diagram
enhancement to improve image clarity. A median filter is The steps involved in the proposed algorithm is
used to suppress noise, followed by histogram equalization to represented as a block diagram.
enhance contrast. These enhancements aid in more accurate
identification of potential nodules in subsequent steps.

IJISRT25APR1725 www.ijisrt.com 3129

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725

Fig 2: Block Diagram of Proposed Algorithm

IV. RESULTS AND DISCUSSION  Enhanced Segmentation: The application of

morphological techniques such as dilation and erosion
To evaluate the effectiveness of the proposed method, a significantly improves the segmentation of lung regions,
dataset from the National Center for Cancer Diseases was leading to more accurate identification of lesions.regions
employed. The performance of the system was measured and nodules, which may not be explicitly handled by other
using a confusion matrix, which provided insights into the deep learning approaches.
model’s accuracy in predicting lung cancer cases. The
method successfully detects and highlights bright regions in  Limitations:
lung images indicative of lesions, enhances image contrast,
and applies morphological analysis to achieve precise  Limited Complexity: The proposed method does not
segmentation. Contour detection is used to define regions of leverage more advanced deep learning architectures like
interest, making the system a powerful tool for early detection U-Net (for precise segmentation), CNN-LSTM (for
of lung cancer. capturing temporal dependencies), or Vision
Transformers (for global context understanding).
The confusion matrix results demonstrate strong  Scalability: The method may not scale as well to larger
performance, with a low number of false positives and false datasets or more complex tasks compared to state-of-the-
negatives. The integration of morphological dilation and art architectures like Vision Transformers.
Sobel edge detection further contributed to improved
accuracy in cancer detection. By comparing predicted  Other Deep Learning Approaches:
classifications with the actual labels of CT scan images, the
system’s diagnostic accuracy was reliably assessed.  U-Net:

A. Comparison with Other Deep Learning Approaches  Strengths: U-Net is specifically designed for medical
image segmentation and excels at tasks requiring precise
 (U-Net, CNN-LSTM, Vision Transformers) localization of objects (e.g., lung nodules). It uses skip
 Proposed Method: CNN + Morphological Analysis connections to combine low-level and high-level features,
making it highly effective for segmentation tasks.
 Strengths:  Comparison: The proposed method may not achieve the
same level of segmentation accuracy as U-Net, which is
 Simplicity: The proposed method is relatively simple and considered a gold standard in medical imaging.
efficient, combining Convolutional Neural Networks with
morphological operations for image segmentation and
classification.

IJISRT25APR1725 www.ijisrt.com 3130

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725
 CNN-LSTM: B. Input Image

 Strengths: CNN-LSTM combines the spatial feature

extraction capabilities of CNNs with the temporal
modeling power of LSTMs. This is useful for tasks
involving sequential data (e.g., video or time-series
medical data).
 Comparison: The proposed method does not incorporate
temporal information, making it less suitable for tasks
requiring sequential analysis.

 Vision Transformers (ViT):

 Strengths: Vision Transformers leverage self-attention

mechanisms to capture global context in images, often
outperforming CNNs in tasks requiring a broad
understanding of the image. Fig 3: Malignant Image
 Comparison: The proposed method relies on CNNs,
which may not capture global context as effectively as
Vision Transformers. ViTs are also more scalable to
larger datasets.

Fig 4: Experimental Results

Fig 5: Prediction

IJISRT25APR1725 www.ijisrt.com 3131

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725
C. Confusion Matrix for Lung Cancer Detection suggested system's dependability was further confirmed by
The model's high accuracy rate suggests that it can the excellent recall and precision metrics.
effectively detect lung cancer. The findings of the confusion
matrix indicated that there were very few false positives and Here is the confusion matrix for lung cancer detection
false negatives, indicating the algorithm's resilience in based on the data you provided. The accuracy of the model is
differentiating between benign and malignant tissues. The approximately 90.32%.

Table 1: Confusion Matrix for Lung Cancer Detection

State Predicted Positive Predicted Negative
True Positive(TP) 26 4
True Negative(TN) 3 2
False Positive(FP) 2 2
False Negative(FN) 1 2

Accuracy = (TP + TN) / (TP + TN + FP + FN)

Fig 6: Confusion Matrix

V. CONCLUSION Characterization of Lung Tumor by Using

Convolution Neural Network. In Journal of Physics:
This work combines CNN, Noise Reduction, Conference Series (Vol. 1717, No. 1, p. 012004). IOP
Morphological processing to provide a novel way of Publishing.
detecting lung lesions. The approach provides substantial [5]. Thakur, S. K., Singh, D. P., & Choudhary, J. (2020).
support for early detection and treatment planning in the Lung cancer identification: a review on detection and
management of lung cancer by increasing the accuracy of classification. Cancer and Metastasis Reviews, 39(3),
lesion segmentation. Future work will focus on expanding the 989-998.
dataset, exploring other deep learning architectures, and [6]. Muthee, M. G., Makau, M., & Amos, O. (2022). A
validating the model on external datasets. review of techniques for morphological analysis in
natural language processing. African Journal of
REFERENCES Science, Technology and Social Sciences, 1(2), 93-
103.
[1]. Liu, X., Song, L., Liu, S., & Zhang, Y. (2021). A [7]. Chen, S., Zhao, M., Wu, G., Yao, C., & Zhang, J.
review of deep-learning-based medical image (2012). Recent advances in morphological cell image
segmentation methods. Sustainability, 13(3), 1224. analysis. Computational and mathematical methods in
[2]. Soni, A., & Singh, A. P. (2020, February). Automatic medicine, 2012(1), 101536.
Pulmonary Cancer Detection using Prewitt & [8]. Ding, C., Aye, H. T. Z., Pa, W. P., Nwet, K. T., Soe,
Morphological Dilation. In 2nd International K. M., Utiyama, M., & Sumita, E. (2019). Towards
Conference on Data, Engineering and Applications Burmese (Myanmar) morphological analysis:
(IDEA) (pp. 1-6). IEEE. Syllable-based tokenization and part-of-speech
[3]. Ahmed, T., Parvin, M. S., Haque, M. R., & Uddin, M. tagging. ACM Transactions on Asian and Low-
S. (2020). Lung cancer detection using CT image Resource Language Information Processing
based on 3D convolutional neural network. Journal of (TALLIP), 19(1), 1-34.
Computer and Communications, 8(03), 35. [9]. Wang, W., Fam, R., Bao, F., Lepage, Y., & Gao, G.
[4]. Sangeetha, M., & Mythili, S. (2021). Detection and (2019, July). Neural morphological segmentation

IJISRT25APR1725 www.ijisrt.com 3132

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1725
model for Mongolian. In 2019 International Joint Thoracic Oncology, 12(3), 501-509.
Conference on Neural Networks (IJCNN) (pp. 1-7). [21]. Biradar, V. G., Pareek, P. K., & Nagarathna, P. (2022,
IEEE. October). Lung Cancer detection and classification
[10]. Ansari, E., Žabokrtský, Z., Mahmoudi, M., using 2D convolutional neural network. In 2022 IEEE
Haghdoost, H., & Vidra, J. (2019, September). 2nd Mysore Sub Section International Conference
Supervised morphological segmentation using rich (MysuruCon) (pp. 1-5). IEEE.
annotated lexicon. In Proceedings of the International [22]. Karim, D. Z., & Bushra, T. A. (2021, December).
Conference on Recent Advances in Natural Language Detecting lung cancer from histopathological images
Processing (RANLP 2019) (pp. 52-61). using convolution neural network. In TENCON 2021-
[11]. Sang, J., Alam, M. S., & Xiang, H. (2019, May). 2021 IEEE Region 10 Conference (TENCON) (pp.
Automated detection and classification for early stage 626-631). IEEE.
lung cancer on CT images using deep learning. [23]. Abdul, W. (2020, December). An automatic lung
In Pattern recognition and tracking XXX (Vol. 10995, cancer detection and classification (ALCDC) system
pp. 200-207). SPIE. using convolutional neural network. In 2020 13th
[12]. Ronneberger, O., Fischer, P., & Brox, T. (2015). U- International Conference on Developments in
net: Convolutional networks for biomedical image eSystems Engineering (DeSE) (pp. 443-446). IEEE.
segmentation. In Medical image computing and [24]. Sasikala, S., Bharathi, M., & Sowmiya, B. R. (2018).
computer-assisted intervention–MICCAI 2015: 18th Lung cancer detection and classification using deep
international conference, Munich, Germany, October CNN. International Journal of Innovative Technology
5-9, 2015, proceedings, part III 18 (pp. 234-241). and Exploring Engineering (IJITEE), 8(2S).
Springer International Publishing. [25]. Perez, G., & Arbelaez, P. (2020). Automated lung
[13]. Soni, A., & Singh, A. P. (2020, February). Automatic cancer diagnosis using three-dimensional
Pulmonary Cancer Detection using Prewitt & convolutional neural networks. Medical & Biological
Morphological Dilation. In 2nd International Engineering & Computing, 58, 1803-1815.
Conference on Data, Engineering and Applications
(IDEA) (pp. 1-6). IEEE.
[14]. Kuruvilla, J., & Gunavathi, K. (2013). Detection of
lung cancer using morphological
operations. International Journal of Scientific &
Engineering Research, 4(8), 1636-1639.
[15]. Hussain, L., Rathore, S., Abbasi, A. A., & Saeed, S.
(2019, March). Automated lung cancer detection
based on multimodal features extracting strategy using
machine learning techniques. In Medical imaging
2019: physics of medical imaging (Vol. 10948, pp.
919-925). SPIE.
[16]. Soni, A., Rai, A., & Shivhare, E. (2019). Atomatic
Lung Cancer Detection using Sobel & Morphological
Operations. International Journal of Engineering and
Advanced Technology (IJEAT).
[17]. Vas, M., & Dessai, A. (2017, August). Lung cancer
detection system using lung CT image processing.
In 2017 international conference on computing,
communication, control and automation
(ICCUBEA) (pp. 1-5). IEEE.
[18]. Kalaivani, S., Chatterjee, P., Juyal, S., & Gupta, R.
(2017, April). Lung cancer detection using digital
image processing and artificial neural networks.
In 2017 International conference of Electronics,
Communication and Aerospace Technology
(ICECA) (Vol. 2, pp. 100-103). IEEE.
[19]. Li, R., Xiao, C., Huang, Y., Hassan, H., & Huang, B.
(2022). Deep learning applications in computed
tomography images for pulmonary nodule detection
and diagnosis: A review. Diagnostics, 12(2), 298.
[20]. Luo, X., Zang, X., Yang, L., Huang, J., Liang, F.,
Rodriguez-Canales, J & Xiao, G. (2017).
Comprehensive computational pathological image
analysis predicts lung cancer prognosis. Journal of

IJISRT25APR1725 www.ijisrt.com 3133