A robust and extendable framework towards fully automated diagnosis of nonmass lesions in breast DCE-MRI

Abstract

Diagnosis of breast nonmass lesions, most notably ductal carcinoma in situ, is challenging. Recent studies show that dynamic contrast enhanced MRI achieves high sensitivity in diagnosis of nonmass lesions. Unlike successfully applied to diagnose mass lesions, particularly kinetic features are reported to be less effective in discriminating nonmass lesions. It is even difficult for human observers to differentiate nonmass lesions against the enhancing parenchymal or benign lesions due to their sometimes similar morphology and contrast kinetics. Towards an automated computer-aided diagnosis system of nonmass lesions, we implemented an extendable and completely automated framework that is efficient and modularized, aiming to discriminate detected suspicious regions into malignant and benign. The entire framework consists of five sequentially executed modules: motion correction, segmentation of breast regions, detection of suspicious regions, feature extraction, and knowledge-based analysis of suspicious regions. A preliminary test was performed on a data set collecting 162 nonmass lesions extracted from 67 patients, which achieved an area under ROC curve value of 0.74 for malignant lesions

Pre and post-hoc diagnosis and interpretation of malignancy from breast DCE-MRI

“…Such automated pre-hoc approaches generally employed an exhaustive search method or clustering to detect ROIs in the scan using hand-designed features (Gubern-Mérida et al, 2015;Mcclymont, 2015;Renz et al, 2012;Wang et al, 2014). The classification of ROIs into false positive, benign or malignant findings is then performed with a new set of hand-designed features extracted from the ROIs (Gubern-Mérida et al, 2015;Mcclymont, 2015;Renz et al, 2012;Wang et al, 2014). These fully automated methods generally suffer from two issues: 1) the sub-optimality of hand-designed features needed at both ROI localization and ROI classification, and 2) the high computational cost of the exhaustive search to detect ROIs.…”

“…Aiming at reducing user intervention to reduce the number of ROIs (Liu et al, 2017), pre-hoc systems evolved to be fully automated. Such automated pre-hoc approaches generally employed an exhaustive search method or clustering to detect ROIs in the scan using hand-designed features (Gubern-Mérida et al, 2015;Mcclymont, 2015;Renz et al, 2012;Wang et al, 2014). The classification of ROIs into false positive, benign or malignant findings is then performed with a new set of hand-designed features extracted from the ROIs (Gubern-Mérida et al, 2015;Mcclymont, 2015;Renz et al, 2012;Wang et al, 2014).…”

Can Occult Invasive Disease in Ductal Carcinoma In Situ Be Predicted Using Computer-extracted Mammographic Features?

“…There has been much work using computer-aided detection (CAD) and diagnosis (CADx) for mammography including microcalcification clusters [14–20]. Other CAD/CADx studies have focused on DCIS [21–24], but those studies have not utilized the diagnostic magnification views routinely available during the workup of suspicious calcifications, which offer additional details not appreciable on routine full-field screening mammographic views. In this work, we hypothesize that computer vision techniques as well as various mammographic features developed for screening detection or diagnosis can be used to help predict the presence of occult invasive disease associated with DCIS.…”