Breast cancer is a major health concern affecting millions of women worldwide. Early detection greatly enhances survival rates and reduces the burden of treatment. This research presents a Computer-Aided Diagnosis (CAD) system that uses advanced image processing and deep learning techniques to classify mammographic images into benign or malignant categories. The system incorporates preprocessing, segmentation using Gaussian Mixture Models (GMM), feature extraction with Gray Level Co-occurrence Matrix (GLCM), and classification using a Probabilistic Neural Network (PNN). Developed in MATLAB and tested on the mini-MIAS database, the system achieved an accuracy of 99.4%, sensitivity of 99.3%, and specificity of 100%. This paper emphasizes the potential of automated systems in supporting medical diagnosis and enabling early intervention.
Introduction
???? Overview
Breast cancer remains a leading cause of cancer-related mortality among women worldwide, with approximately 2.3 million new cases annually. Early detection is crucial for effective treatment, especially in resource-limited settings. Traditional diagnostic methods like biopsies and manual screenings are often delayed and subjective. This research introduces an automated diagnostic tool leveraging PNN and GLCM for texture-based classification of mammographic images.
???? Methodology
Dataset: Utilized the mini-MIAS database, comprising 322 labeled mammogram images categorized as normal, benign, or malignant.
Preprocessing:
Converted images to grayscale and resized to 1024×1024 pixels.
Applied Adaptive Median Filtering to remove noise.
Enhanced contrast using Histogram Modified CLAHE.
Used morphological dilation to highlight microcalcification clusters.
Segmentation:
Employed Gaussian Mixture Model (GMM)-based segmentation to delineate regions of interest.
Applied Otsu’s thresholding and top-hat morphological operations to isolate microcalcifications and remove uneven illumination.
Feature Extraction:
Extracted texture features using GLCM at four orientations (0°, 45°, 90°, 135°), including contrast, energy, entropy, homogeneity, and correlation.
Classification:
Implemented PNN for classification due to its rapid training and high accuracy.
PNN utilized a Parzen window estimator with a Gaussian kernel to compute class probabilities.
???? System Implementation
Software: MATLAB 2018a with GUIDE for GUI development.
Hardware:
Processor: Intel Core i5 (Dual Core)
RAM: 4 GB DDR III
Storage: 500 GB HDD
Monitor: 15” LCD
OS: Windows 10
Training:
Employed data augmentation techniques such as rotation, mirroring, and flipping to address class imbalance.
70% of images used for training, 30% for testing.
Training time was approximately 5 minutes using 100 epochs.
User Interface:
Developed a GUI for image upload, real-time classification, and result visualization, highlighting cancerous regions.
???? Results
Accuracy: 99.4%
Sensitivity: 99.3%
Specificity: 100%
Precision: 98.9%
F1 Score: 99.1%
Processing Time per Image: ~1.2 seconds
These results demonstrate the efficacy of combining GLCM and PNN, with the system effectively handling noisy inputs and varying image quality.
???? Performance Comparison
Method
Accuracy
Sensitivity
Specificity
SVM + GLCM
92.1%
89.5%
93.4%
CNN
95.7%
94.1%
96.2%
Proposed (PNN + GLCM)
99.4%
99.3%
100%
?? Limitations
Dataset Size: The system was trained on a small dataset; performance on larger, real-world clinical datasets may vary.
Feature Scope: Only texture-based features were utilized; incorporating shape and edge-based features could further enhance accuracy.
Platform Constraints: The MATLAB-based interface may limit real-time deployment on mobile devices or embedded systems.
Conclusion
This research successfully demonstrates the design and implementation of a CAD system using PNN and GLCM, capable of classifying breast cancer mammograms with high accuracy. The combination of adaptive preprocessing, texture-based feature extraction, and PNN classification leads to reliable diagnostic outcomes.
References
[1] Kourou, K., et al. “Machine learning applications in cancer prognosis and prediction.” Comput. Struct. Biotechnol. J., 2015.
[2] Spanhol, F. A., et al. “A dataset for breast cancer histopathological image classification.” IEEE Trans. Biomed. Eng., 2016.
[3] Sreedevi, T., et al. “Breast cancer detection using GLCM and PNN classifier.” Int. J. Adv. Comput. Sci., 2020.
[4] Charu C. Aggarwal, Data Classification Algorithms and Applications, CRC Press, 2015, pp. 4–64.
[5] Mohapatra, P. and Yegnanarayana, B. “Mammogram classification using texture features,” in ICIP, 2012.
[6] Cruz-Roa, A., et al. “CNNs for breast cancer histology image analysis,” 2014.
[7] Dai, Q., Zhang, C., and Wu, H. \"Research of Decision Tree Classification Algorithm in Data Mining\", Int. J. Database Theory Appl., vol. 9, no. 5, 2016, pp. 1–8.
[8] Joshi, K.K. and Patidar, P. \"Construct Decision Tree using Sliq and KNN\", ISSN: 2277-5528, pp. 231–236.
[9] Ladha, L. and Deepa, T. \"Feature Selection Methods\", Int. J. Comput. Sci. Eng., vol. 3, 2011, pp. 1787–1797.
[10] Patel, H. and Patel, D. \"Survey on Data Mining for Agriculture\", Int. J. Comput. Appl., vol. 95, no. 9, 2014.
[11] Patel, B.R. and Rana, K.K. \"Survey on Decision Tree for Classification\", Int. J. Eng. Dev. Res., vol. 2, 2014, pp. 1–5.
[12] Patel, B.N., Prajapati, S.G. and Lakhtaria, K.I. \"Efficient Data Classification using Decision Tree\", Bonfring Int. J. Data Mining, vol. 2, no. 1, 2012, pp. 6–12.
[13] www.breastcancer.org
[14] www.geeksforgeeks.org
[15] www.mathworks.com/products/matlab.html
[16] https://www.niddk.nih.gov/.../breastcancer-disease-stroke
[17] https://www.breastcancer.org/.../cardiovascular-disease--breastcancer
[18] https://www.geeksforgeeks.org/naive-bayes-classifiers/
[19] https://towardsdatascience.com/support-vector-machine-introduction...
[20] www.towardsdatascience.com