Breast Tumor Segmentation and Shape Classification in Mammograms using Generative Adversarial and Convolutional Neural Network (1809.01687v3)

Abstract: Mammogram inspection in search of breast tumors is a tough assignment that radiologists must carry out frequently. Therefore, image analysis methods are needed for the detection and delineation of breast masses, which portray crucial morphological information that will support reliable diagnosis. In this paper, we proposed a conditional Generative Adversarial Network (cGAN) devised to segment a breast mass within a region of interest (ROI) in a mammogram. The generative network learns to recognize the breast mass area and to create the binary mask that outlines the breast mass. In turn, the adversarial network learns to distinguish between real (ground truth) and synthetic segmentations, thus enforcing the generative network to create binary masks as realistic as possible. The cGAN works well even when the number of training samples are limited. Therefore, the proposed method outperforms several state-of-the-art approaches. This hypothesis is corroborated by diverse experiments performed on two datasets, the public INbreast and a private in-house dataset. The proposed segmentation model provides a high Dice coefficient and Intersection over Union (IoU) of 94% and 87%, respectively. In addition, a shape descriptor based on a Convolutional Neural Network (CNN) is proposed to classify the generated masks into four mass shapes: irregular, lobular, oval and round. The proposed shape descriptor was trained on Digital Database for Screening Mammography (DDSM) yielding an overall accuracy of 80%, which outperforms the current state-of-the-art.

• The paper introduces a novel deep learning framework using cGAN for accurate breast tumor segmentation, achieving a Dice coefficient of 94% and an IoU of 87%.
• The paper demonstrates the effectiveness of a CNN-based shape descriptor for classifying tumors into irregular, lobular, oval, and round with an overall accuracy of 80%.
• The paper highlights significant advancements for CAD systems in medical imaging while suggesting future research directions to improve diagnostic robustness and expand segmentation capabilities.

Overview of Segmentation and Shape Classification Techniques in Mammograms

The analyzed paper introduces a novel framework for breast tumor segmentation and shape classification using both Conditional Generative Adversarial Networks (cGAN) and Convolutional Neural Networks (CNN). This research focuses on improving the accuracy and efficiency of identifying and classifying tumors within mammograms by leveraging advanced deep learning techniques.

The central contribution of the paper is the use of a cGAN to segment breast tumors in mammograms, highlighting the model's ability to perform well even with limited training samples. The generative network of the cGAN accurately recognizes tumor areas and generates corresponding binary masks. The adversarial network, meanwhile, distinguishes between ground truth and synthetic segmentations, pushing the generative network towards producing more realistic masks. The segmentation model achieves a Dice coefficient of up to 94% and an Intersection over Union (IoU) of approximately 87% on two datasets—the public INbreast and a private in-house dataset.

Additionally, the research proposes a CNN-based shape descriptor for classifying tumor shapes, achieving an overall accuracy of 80% on the Digital Database for Screening Mammography (DDSM). Classifications are made into four primary tumor shapes: irregular, lobular, oval, and round. This marks an improvement over current state-of-the-art methods in both segmentation and classification tasks, showcasing the potential of deep learning models in medical image analysis.

Key Numerical Results and Comparative Analysis

The research presents significant improvements in both the segmentation and classification tasks compared to existing approaches. The proposed framework's segmentation model showed a marked increase in accuracy with metrics such as Dice coefficients reaching 94% and IoUs being 87%, outperforming several existing state-of-the-art algorithms. Such results are indicative of the granularity and precision the cGAN model brings to segmenting breast tumors.

In shape classification, leveraging the CNN model attained a notable 80% accuracy, illustrating its efficacy over previous methodologies, especially in differentiating between intricate tumor shapes. The superior classification accuracy achieved by focusing on the morphological aspects using CNN is compelling, given the complexity entailed in distinguishing tumor shapes.

Implications and Future Prospects

The implications of this research are substantial for the domain of medical imaging, particularly in enhancing Computer-Aided Diagnosis (CAD) systems which are crucial for aiding radiologists in precise and efficient breast cancer diagnosis. The combination of cGAN with CNN enhances both segmentation and classification tasks, rightfully positioning this technique as an effective tool in the diagnostic process.

Future prospects could involve testing the framework on larger and more diverse datasets to validate its robustness. Further explorations could include extending the framework to segment and classify other forms of tumors or additional mammographic features like microcalcifications or architectural distortions. Such enhancements could holistically improve diagnostic capabilities and potentially integrate other features like tumor margin analysis into the current system.

Additionally, researchers might investigate more efficient architectures or optimizers to further increase the system’s accuracy and reduce computational load. The demonstrated capability of cGAN to yield impressive results with limited data suggests rich avenues for exploring similar paradigms across diverse applications in medical image processing.

In conclusion, this paper contributes valuable insights and advancements within the medical imaging domain, suggesting practical applications and future explorations that could considerably impact clinical practices and CAD systems.