Medical Imaging Technology

Permanent URI for this collection

http://dspace.alquds.edu:4000/handle/20.500.12213/6128

Browse

Browsing Medical Imaging Technology by Author "عمر فايق صادق دراغمه"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    A Hybrid Artificial Intelligence Approach for Early Detection of Breast Cancer and Classification from Mammogram Images in Palestine
    (Al-Quds University, 2024-05-26) Omar Faiq Sadeq Daraghmeh; عمر فايق صادق دراغمه
    Breast cancer is a significant global health concern, especially in Palestine, and early and accurate diagnosis is crucial for improving patient outcomes and survival rates. However, despite advancements in medical technology and screening techniques, missed diagnoses remain a persistent challenge in breast cancer detection. This study investigates the use of hybrid artificial intelligence (AI) models that combine deep learning and machine learning techniques to predict benign and malignant breast cancer from mammogram images. The study starts by utilizing pre-trained convolutional neural network models, namely VGG16 and DenseNet121, for feature extraction from mammogram images. These deep learning models have been trained on large datasets and have learned to identify various patterns and features within images. By extracting these features from mammograms, the models can capture important information that is relevant to the classification of breast cancer. The extracted features are then used to train several machine learning classifiers, including logistic regression, support vector machines, random forests, and gradient boosting models. These classifiers learn to recognize patterns and make predictions based on the extracted features. To evaluate the performance of the hybrid AI models, the study is conducted in three stages. In the first stage, the original mammogram images are used for classification. In the second stage, the mammogram images are enhanced using various image preprocessing and enhancement techniques. Finally, in the third stage, the models are tested on new mammogram images to assess their generalization capabilities. To enhance the mammogram images, several image processing techniques are applied. These include morphological erosion preprocessing, Contrast-Limited Adaptive Histogram Equalization (CLAHE), Laplacian of Gaussian (LoG) edge enhancement, and unsharp masking. These techniques aim to improve the visibility of important structures and features within the images, making it easier for the AI models to make accurate predictions. In the second stage, when predicting benign cases from the enhanced mammogram images, the logistic regression classifier with DenseNet121 features achieves remarkable performance. It achieves the highest accuracy of 0.991, precision of 0.996, F1-score of 0.989, and an AUC of 0.999. The support vector machine with DenseNet121 features also performs well, with an accuracy of 0.986 and an AUC of 0.999. The logistic regression model with VGG16 features demonstrates the fastest predictive time, requiring only 0.13 seconds. Similarly, in predicting malignant cases from the enhanced images, the logistic regression classifier with DenseNet121 features excels with the highest accuracy of 0.995, precision of 0.995, recall of 0.995, F1-score of 0.995, and an AUC of 0.999. The support vector machine with DenseNet121 features follows closely with an accuracy of 0.992 and an AUC of 0.998. The logistic regression model with VGG16 features maintains its fast predictive time, taking only 0.08 seconds. The study demonstrates that the enhanced mammogram images in the second stage consistently outperform the original and new test images in the first and third stages, respectively. This emphasizes the significant impact of image preprocessing and enhancement techniques on the predictive capabilities of the hybrid AI models. The findings highlight the potential of combining deep learning for feature extraction and machine learning for classification in achieving high accuracy, precision, recall, F1-scores, and AUC values for predicting breast cancer malignancy from mammogram images. In conclusion, the study demonstrates the potential of hybrid AI models that combine deep learning and machine learning techniques for the prediction of benign and malignant breast cancer from mammogram images. The integration of deep learning for feature extraction and machine learning for classification, along with image preprocessing and enhancement, results in improved accuracy and performance. These advancements have the potential to enhance breast cancer detection, ultimately leading to better patient outcomes and survival rates.