Advancing mammography breast mass detection through diffusion segmentation
Guerroudji, Mohamed Amine,Amara, Kahina,Zenati, Nadia
DOI: https://doi.org/10.1007/s11042-024-18840-6
IF: 2.577
2024-03-28
Multimedia Tools and Applications
Abstract:Medicine has become a necessary component of our daily life in the modern world. In this ever-changing environment, computer-aided diagnosis (CAD) has evolved as a dynamic and essential topic, providing crucial assistance to medical practitioners in their diagnostic endeavors. The main goal of this project is to create a CAD system that provides radiologists with a trustworthy second opinion during comprehensive detection of downscaled MIAS data. Our innovative approach is based on anisotropic denoising and the use of an alternating sequential filter (ASF) to identify potential mass sites. The method begins with ASF extraction through mathematical morphology, paving the way for overall contrast enhancement. To achieve this crucial aspect of our work, we use image-based active geometric contour models (level set) for spot segmentation. Our proposed strategy enables significant image improvements, leading to an increase in detection accuracy and efficient extraction of masses from the mini-MIAS mammography dataset. Experimental findings demonstrated that, when compared to contemporary methods, the proposed method produces satisfactory outcomes. The proposed technique achieved a sensitivity of 93.2%, an accuracy of 97.09%, a Dice coefficient (F1-score) exceeding 95.32%, and a specificity of 97.42%.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering