Plně integrovaný systém podpory rozhodování pro detekci a segmentaci lézí na digitálním mamografu

KVAK, Daniel. Plně integrovaný systém podpory rozhodování pro detekci a segmentaci lézí na digitálním mamografu. In Medsoft 2022. 2022.

Základní údaje

Originální název

Plně integrovaný systém podpory rozhodování pro detekci a segmentaci lézí na digitálním mamografu

Název anglicky

Fully Integrated Decision-Support System for Detection and Segmentation of Breast Lesions in Digital Mammogram

Autoři

KVAK, Daniel

Vydání

Medsoft 2022, 2022

---

Další údaje

Typ výsledku

Prezentace na konferencích

Utajení

není předmětem státního či obchodního tajemství

Odkazy

URL

Klíčová slova anglicky

reast Cancer, Breast Lesion Detection, Computer-Aided Detection, Deep Learning, Image Segmentation, Mammogram, Picture Archiving and Communicating System

Příznaky

Recenzováno

---

Anotace

V originále

Breast cancer is one of the most prevalent forms of cancer affecting women. Detection of suspicious lesions on mammographic images is considered a challenging task due to the variability of lesion sizes and shapes, the problematic margins of the findings, and some extremely small lesions that are difficult to localize. With the increasing availability of digitized clinical archives and the development of complex deep learning (DL) methods, we are witnessing a trend towards the integration of robust computer-aided detection (CAD) systems to assist in the automatic segmentation of lesions on mammograms to aid in the diagnosis of breast cancer. This study presents deep learning–based automatic detection algorithm (DLAD), directly implemented in picture archiving and communication system (PACS) to aid in improving the radiologist's workflow. The proposed DLAD is evaluated on INbreast dataset with a sample size of n=138 (71 [51.45\%] BI-RADS 4/5/6 images, 67 [48.55\%] BI-RADS 1 images). Preliminary results show a sensitivity of 0.9296 [95\% CI 0.8701-0.9891], specificity of 0.7273 [0.6207-0.8339] and IoU of 0.5661, indicating a low false negative rate while maintaining a reasonable false positive rate.

Anglicky

Breast cancer is one of the most prevalent forms of cancer affecting women. Detection of suspicious lesions on mammographic images is considered a challenging task due to the variability of lesion sizes and shapes, the problematic margins of the findings, and some extremely small lesions that are difficult to localize. With the increasing availability of digitized clinical archives and the development of complex deep learning (DL) methods, we are witnessing a trend towards the integration of robust computer-aided detection (CAD) systems to assist in the automatic segmentation of lesions on mammograms to aid in the diagnosis of breast cancer. This study presents deep learning–based automatic detection algorithm (DLAD), directly implemented in picture archiving and communication system (PACS) to aid in improving the radiologist's workflow. The proposed DLAD is evaluated on INbreast dataset with a sample size of n=138 (71 [51.45\%] BI-RADS 4/5/6 images, 67 [48.55\%] BI-RADS 1 images). Preliminary results show a sensitivity of 0.9296 [95\% CI 0.8701-0.9891], specificity of 0.7273 [0.6207-0.8339] and IoU of 0.5661, indicating a low false negative rate while maintaining a reasonable false positive rate.