Error investigation of models for improved detection of masses in screening mammography

Eltonsy, Nevine H.; Essock-Burns, E.; Tourrasi, G.; Elmaghraby, Adel

doi:10.1109/isspit.2005.1577200

Cited by 2 publications

(3 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The technique first targets focal lesions of potential malignancies based on concentric morphology presented around the focal layer and the relative incidence of the focal layer depicted by morphological characteristics. In the development stage, empirical optimization of the scheme's critical parameters was performed based on 250 CC contained 109 malignant masses randomly chosen from the same database [12]. The study compared the performance of training using different linear and nonlinear discriminant models.…”

Section: Resultsmentioning

confidence: 99%

“…The study showed that the trained CAD scheme performance for masses can maintain 99.08% (108/109) and 96% (105/109) sensitivity with false positive rate (FPI) of 8 and 5.6 per image. These preliminary results suggested that further analysis is needed to reduce FPI for the CAD algorithm to be more clinically useful [12]. The document is organized as follows.…”

Section: Resultsmentioning

confidence: 99%

“…Furthermore, the smoothness of the lesion boundary, a crucial diagnostic indicator (s. 3.3), would depend directly on the chosen threshold. These requirements were accounted for a modified 3D pyramidlinking approach [11,12], where a region based segmentation algorithm with threshold independence was devised. Based on a hierarchical Gauß-pyramid, the algorithm fulfils split and merge operations.…”

Section: Contrast Enhancement Segmentationmentioning

confidence: 99%

See 2 more Smart Citations

Special Session on Breast CAD

Chang

Huang²,

Lee

et al. 2006

Int J CARS

View full text Add to dashboard Cite

A fully automatic prototype software system for cancer detection and diagnosis on dynamic contrast enhanced MR investigations of the female breast has been developed The software implements an image processing pipeline including image registration, segmentation, calculation of dynamic and morphologic features and feature classification by Artificial Neural Networks. Without the need of any user-interaction the software generates a diagnosis output that may be used as second opinion. Keywords Registration AE Segmentation AE Pyramid Linking AE Neuronal Networks AE MRI AE Mammography 1 Introduction The diagnosis of breast cancer in contrast enhanced MRI is demanding and time consuming. The evaluation is complicated by motion artefacts, false positive contrast enhancing foci and the high amount of images. To reach a diagnosis, the physician has to identify first and then assess several morphological and dynamic features of each cancer suspicious lesion, all these tasks show a high inter-observer variability. In order to standardise this examination and to help improve the overall performance by a computer generated second opinion, we have developed a fully automatic prototype software for the Computer Aided Detection and Diagnosis (CAD 2 ) of breast lesions. 2 Related work Several groups reported various rigid or non-rigid image registration algorithms for reducing motion artefacts of contrast enhanced MRI of the breast [1]. Independently, different types of Artificial Neural Networks (ANN) have been used for lesion detection [2, 3] as well as characterization of dynamic features [4, 5]. In a previous study ANN proved to be superior to statistical means in evaluation of dynamic and morphological features of contrast enhanced lesions [6]. Up to now, an integration of all different means is -to our knowledge -not published in literature. MethodsAll examinations of the female breast have been performed on a 1.5T MR Scanner (Siemens Magnetom SonataÒ). Each dynamic MRI study consists out of five consecutively measured 3D-series with a acquisition time of 72 seconds [7]. First, a pre-contrast series is acquired, than a contrast media was injected for enhancing contrast between normal glandular tissue and suspicious lesions (CE-MRI), thereafter the post-contrast images series are obtained. Each pre-and post-contrast 3D-series consists of 512 · 256 · 80 voxels with a spatial resolution of 0.7 · 0.7 · 1.8 mm. At the end of an investigation, the pre-contrast series is subtracted from all post-contrast image series. As a result signal intensities of all non enhanced structures of the breast are neutralised, while suspicious tissue or potential lesions are strongly emphasized caused by contrast media uptake in the resulting subtraction series. To perform a fully automatic CAD 2 of the contrast enhanced regions a software prototype was engineered. It executes a pipeline of different image processing operations reproducing the diagnostic approach of a physician and resulting in a suggested diagnosis for each lesion found. 3.1 Moti...

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Contrast Enhancement Segmentationmentioning

confidence: 99%

See 1 more Smart Citation