Enhanced Detection of Breast Cancer Microcalcifications in MRI via Hybrid LOOP and Haralick Features with Stacked Autoencoder Classification
Year: 2025
Type: article
Abstract: Breast cancer is a prevalent and life-threatening disease among women worldwide, where early detection is crucial for effective treatment. Magnetic Resonance Imaging (MRI) is widely utilized for identifying breast tumors due to its superior sensitivity and spatial resolution. However, accurately detecting microcalcifications—small calcium deposits linked to malignancy—remains a significant challenge with conventional imaging analysis techniques like Haralick texture features alone. To overcome these limitations, this study introduces a novel hybrid feature extraction method that combines Local Optimal Oriented Patterns (LOOP) with Haralick texture descriptors, enhancing the representation of microcalcifications in breast MRI scans. The proposed method includes preprocessing the MRI images using Contrast Limited Adaptive Histogram Equalization (CLAHE) to reduce noise, followed by segmentation performed through morphological operations, multilevel Otsu thresholding, and region-growing approaches to isolate relevant breast tissues. Extracted hybrid features are then input into a Stacked Autoencoder-based Deep Neural Network (DNN) classifier to differentiate between malignant and benign cases. An experimental evaluation was conducted on a BI-RADS MRI dataset comprising over 900 clinical breast images. The hybrid feature extraction approach demonstrated improved diagnostic performance compared to traditional methods, achieving an accuracy of 95.33%, a sensitivity of 95.81%, a specificity of 94.67%, a positive predictive value (PPV) of 94.82%, and a negative predictive value (NPV) of 94.55%. These results represent a 3.83% accuracy increase over baseline classifiers such as Support Vector Machines (SVM), K-Nearest Neighbours (KNN), and Random Forests (RF). Our findings confirm that integrating LOOP with Haralick features significantly enhances breast cancer detection from MRI images, thereby providing a robust tool to assist clinicians in early diagnosis and improving patient outcomes.
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7160964178