A Novel Deep Learning Framework for Nipple Segmentation in Digital Mammography DOI
Marcos Rogozinski, Jan Hurtado, César A. Sierra-Franco

и другие.

Deleted Journal, Год журнала: 2025, Номер unknown

Опубликована: Июнь 3, 2025

This study introduces a novel methodology to enhance nipple segmentation in digital mammography, critical component for accurate medical analysis and computer-aided detection systems. The is key anatomical landmark multi-view multi-modality breast image registration, where localization vital ensuring quality enabling precise registration of anomalies across different mammographic views. proposed approach significantly outperforms baseline methods, particularly challenging cases previous techniques failed. It achieved successful all reached mean Intersection over Union (mIoU) 0.63 instances the failed entirely. Additionally, it yielded nearly tenfold improvement Hausdorff distance consistent gains overlap-based metrics, with mIoU increasing from 0.7408 0.8011 craniocaudal (CC) view 0.7488 0.7767 mediolateral oblique (MLO) view. Furthermore, its generalizability suggests potential application other imaging modalities related domains facing challenges such as class imbalance high variability object characteristics.

Язык: Английский

Black-box Adversarial Attack Defense Approach: An Empirical Analysis from Cybersecurity Perceptive DOI Creative Commons
Kousik Barik, Sanjay Misra, Inés López-Baldominos

и другие.

Results in Engineering, Год журнала: 2025, Номер unknown, С. 105177 - 105177

Опубликована: Май 1, 2025

Язык: Английский

Процитировано

1
A Novel Deep Learning Framework for Nipple Segmentation in Digital Mammography DOI
Marcos Rogozinski, Jan Hurtado, César A. Sierra-Franco

и другие.

Deleted Journal, Год журнала: 2025, Номер unknown

Опубликована: Июнь 3, 2025

This study introduces a novel methodology to enhance nipple segmentation in digital mammography, critical component for accurate medical analysis and computer-aided detection systems. The is key anatomical landmark multi-view multi-modality breast image registration, where localization vital ensuring quality enabling precise registration of anomalies across different mammographic views. proposed approach significantly outperforms baseline methods, particularly challenging cases previous techniques failed. It achieved successful all reached mean Intersection over Union (mIoU) 0.63 instances the failed entirely. Additionally, it yielded nearly tenfold improvement Hausdorff distance consistent gains overlap-based metrics, with mIoU increasing from 0.7408 0.8011 craniocaudal (CC) view 0.7488 0.7767 mediolateral oblique (MLO) view. Furthermore, its generalizability suggests potential application other imaging modalities related domains facing challenges such as class imbalance high variability object characteristics.

Язык: Английский

Процитировано

0