Cellular Neural Networks and Development of a Decision Support Systems for Breast Cancer Detection  

Tarih: 10.03.2015
Yer: Kandilli Kampüs, AZ-19 13:00- 14:30

Halil Özcan GÜLÇÜR  Boğaziçi Üniversitesi, BME

MRM is gaining increased acceptance for detecting breast cancer in its early stages since it offers the highest sensitivity among other breast imaging modalities. In MRM, lesion-obscuring overlapping structures and summation shadows are much less pronounced as there is no need for excessive breast compression during imaging. Moreover, it does not use harmful ionizing radiation and with the help of contrast agents, it provides important tissue information on cross-sectional morphology, as well as functional information on perfusion and capillary leakage.

For accurate assessment of breast cancer however, a large volume of image data must be acquired at high spatial and temporal resolutions with an MR scanner and must be analyzed meticulously. The morphology and enhancement dynamics of every suspicious region must be visually evaluated diligently by placing a small region of interest (ROI) over the early enhancing component of the lesion and tracking its course over time. As these regions can be very small and there may be several deceptively enhanced regions from healthy tissues, manual identification of lesions from acquired images and subtraction images requires intensive attention of an expert radiologist for periods that may exceed 30 min. Moreover, the final diagnoses reflect the experience of the examiners and are thus subjective.

DynaMammoAnalyst is a novel decision support software developed by the authors to speed up the examinations and overcome subjectiveness via computation of several enhancement parameters. It facilitates lesion identification, delineation and evaluation by providing improved visualization, segmentation and localization of suspiciously enhancing regions, interactive plots of time–intensity curves and time–intensity curve distributions. It shortens the time required to explore breast MRI data considerably and reduces inter- and intra-observer variability by providing decision support for simultaneous quantitative dynamic and morphologic evaluations. DynaMammoAnalyst is able to process breast MR images acquired in different orientations with different resolutions and is able to show images in DICOM format from other modalities such as X-ray mam- mography and breast ultrasound, supplying adjunct diagnostic information to the radiologist. It requires minimal user interaction and post-processing of the images is pleasingly fast.

DynaMammoAnalyst uses Cellular Neural Networks (CNNs) which are massively parallel cellular structures with learning abilities to realize complex image processing tasks efficiently and in almost real time. In this preliminary study, we propose a novel, robust, and fully automated system based on CNNs to facilitate lesion localization in contrast-enhanced MR mammography, a difficult task requiring the processing of a large number of images with attention paid to minute details. The data set consists of  slices containing precontrast and postcontrast bilateral axial MR mammograms. The breast region of interest is first segmented from the precontrast images using four 2-D CNNs connected in cascade, specially designed to minimize false detections due to muscles, heart, lungs, and thoracic cavity. To identify deceptively enhancing regions, a 3-D nMITR map of the segmented breast is computed and converted into binary form. During this process tissues that have low degrees of enhancements are discarded. To boost lesions, this binary image is processed by a 3D CNN with a control template consisting of three layers of 11 X 11 cells and a fuzzy c-partitioning output function. A set of decision rules extracted empirically from the training data set based on volume and 3D eccentricity features is used to make final decisions and localize lesions. The segmentation algorithm performs well with high average precision, high true positive volume fraction, and low false positive volume fraction with an overall performance and the lesion detection performance of the system is quite satisfactory.

In the seminar  the design philosophy of decision support systems and cellular neural networks will also  be explained.