TY - JOUR
T1 - Development and Functional Validation of A Software System for Detecting and Classifying Artifacts in Mammographic Images
AU - Franco Arias, Manuel Hernando
A2 - Ariza Carrillo, Diego Andres
A2 - Martinez Malaver, Jennifer Andrea
A2 - Arboleda Carvajal, Alejandro
A2 - Morales Cordero, Mario Fernando
A2 - Morales Salcedo, Angela Patricia
PY - 2026/9/16
Y1 - 2026/9/16
N2 - Mammography is a fundamental imaging modality for breast cancer detection and diagnosis; however, its diagnostic performance can be compromised by image artifacts that obscure relevant anatomical information, simulate findings, and may lead to repeated examinations. The automated identification of such artifacts remains less explored than computational approaches for lesion detection and classification, creating a need for dedicated tools for mammographic quality control. This study aimed to develop and functionally validate MEFACT, a software system designed to detect and classify contrast splatter, white-pixel defects, and radiolucent-line artifacts in mammographic images. The system was implemented in Python using artifact-specific image-processing strategies combined with convolutional neural network classification for contrast splatter and radiolucent lines, while white-pixel defects were detected through a pixel-wise routine based on local intensity changes, inversion, and dilation. Functional testing included 45 images from each artifact category and 45 artifact-free mammograms. The contrast-splatter and radiolucent-line networks achieved training accuracies of 88.5% and 99.35%, respectively. During functional testing, the system correctly identified 44 of 45 contrast-splatter images, 45 of 45 white-pixel images, 43 of 45 radiolucent-line images, and all 45 artifact-free mammograms, corresponding to 177 correct decisions among 180 evaluated images. Statistical comparisons with artifact-free mammograms showed significant differences in system outputs. These findings demonstrate the potential of MEFACT as a software-assisted tool for mammographic quality control, while emphasizing the need for larger and more diverse image datasets to improve reliability and discrimination between artifact categories.
AB - Mammography is a fundamental imaging modality for breast cancer detection and diagnosis; however, its diagnostic performance can be compromised by image artifacts that obscure relevant anatomical information, simulate findings, and may lead to repeated examinations. The automated identification of such artifacts remains less explored than computational approaches for lesion detection and classification, creating a need for dedicated tools for mammographic quality control. This study aimed to develop and functionally validate MEFACT, a software system designed to detect and classify contrast splatter, white-pixel defects, and radiolucent-line artifacts in mammographic images. The system was implemented in Python using artifact-specific image-processing strategies combined with convolutional neural network classification for contrast splatter and radiolucent lines, while white-pixel defects were detected through a pixel-wise routine based on local intensity changes, inversion, and dilation. Functional testing included 45 images from each artifact category and 45 artifact-free mammograms. The contrast-splatter and radiolucent-line networks achieved training accuracies of 88.5% and 99.35%, respectively. During functional testing, the system correctly identified 44 of 45 contrast-splatter images, 45 of 45 white-pixel images, 43 of 45 radiolucent-line images, and all 45 artifact-free mammograms, corresponding to 177 correct decisions among 180 evaluated images. Statistical comparisons with artifact-free mammograms showed significant differences in system outputs. These findings demonstrate the potential of MEFACT as a software-assisted tool for mammographic quality control, while emphasizing the need for larger and more diverse image datasets to improve reliability and discrimination between artifact categories.
M3 - Artículo Científico
SN - 3079-0875
VL - 3
SP - 1015
EP - 1023
JO - Journal of Intelligent Decision Making and Information Science
JF - Journal of Intelligent Decision Making and Information Science
IS - 12s
ER -