S. Tangaro

Mammogram segmentation by contour searching and massive lesion classification with neural network

The mammography is the most effective procedure for an early diagnosis of the breast cancer. In this paper, an algorithm for detecting massive lesions in mammographic images will be presented. The database consists of 3762 digital images acquired in several hospitals belonging to the MAGIC-5 collaboration. A reduction of the surface under investigation is achieved, without loss of meaningful information, through segmentation of the whole image, by means of a ROI Hunter algorithm. In the following classification step, feature extraction plays a fundamental role: some features give geometrical information, other ones provide shape parameters. Once the features are computed for each ROI, they are used as inputs to a supervised neural network with momentum. The output neuron provides the probability that the ROI is pathological or not. Results are provided in terms of ROC and FROC curves; the area under the ROC curve was found to be Az=(85.6plusmn0.8)%. This software is included in the CAD station actually working in the hospitals belonging to the MAGIC-5 Collaboration

A completely automated CAD system for mass detection in a large mammographic database

Mass localization plays a crucial role in computer-aided detection (CAD) systems for the classification of suspicious regions in mammograms. In this article we present a completely automated classification system for the detection of masses in digitized mammographic images. The tool system we discuss consists in three processing levels: (a) Image segmentation for the localization of regions of interest (ROIs). This step relies on an iterative dynamical threshold algorithm able to select iso-intensity closed contours around gray level maxima of the mammogram. (b) ROI characterization by means of textural features computed from the gray tone spatial dependence matrix (GTSDM), containing second-order spatial statistics information on the pixel gray level intensity. As the images under study were recorded in different centers and with different machine settings, eight GTSDM features were selected so as to be invariant under monotonic transformation. In this way, the images do not need to be normalized, as the adopted features depend on the texture only, rather than on the gray tone levels, too. (c) ROI classification by means of a neural network, with supervision provided by the radiologists diagnosis. The CAD system was evaluated on a large database of 3369 mammographic images [2307 negative, 1062 pathological (or positive), containing at least one confirmed mass, as diagnosed by an expert radiologist]. To assess the performance of the system, receiver operating characteristic (ROC) and free-response ROC analysis were employed. The area under the ROC curve was found to be Az = 0.783 +/- 0.008 for the ROI-based classification. When evaluating the accuracy of the CAD against the radiologist-drawn boundaries, 4.23 false positives per image are found at 80% of mass sensitivity.

A CAD system for nodule detection in low‐dose lung CTs based on region growing and a new active contour model

A computer-aided detection (CAD) system for the selection of lung nodules in computer tomography (CT) images is presented. The system is based on region growing (RG) algorithms and a new active contour model (ACM), implementing a local convex hull, able to draw the correct contour of the lung parenchyma and to include the pleural nodules. The CAD consists of three steps: (1) the lung parenchymal volume is segmented by means of a RG algorithm; the pleural nodules are included through the new ACM technique; (2) a RG algorithm is iteratively applied to the previously segmented volume in order to detect the candidate nodules; (3) a double-threshold cut and a neural network are applied to reduce the false positives (FPs). After having set the parameters on a clinical CT, the system works on whole scans, without the need for any manual selection. The CT database was recorded at the Pisa center of the ITALUNG-CT trial, the first Italian randomized controlled trial for the screening of the lung cancer. The detection rate of the system is 88.5% with 6.6 FPs/CT on 15 CT scans (about 4700 sectional images) with 26 nodules: 15 internal and 11 pleural. A reduction to 2.47 FPs/CT is achieved at 80% efficiency.

Direct analysis of molybdenum target generated x-ray spectra with a portable device

In routine applications, information about the photon flux of x-ray tubes is obtained from exposure measurements and cataloged spectra. This approach relies mainly on the assumption that the real spectrum is correctly approximated by the cataloged one, once the main characteristics of the tube such as voltage, target material, anode angle, and filters are taken account of. In practice, all this information is not always available. Moreover, x-ray tubes with the same characteristics may have different spectra. We describe an apparatus that should be useful for quality control in hospitals and for characterizing new radiographic systems. The apparatus analyzes the spectrum generated by an x-ray mammographic unit. It is based on a commercial CZT produced by AMPTEK Inc. and a set of tungsten collimator disks. The electronics of the CZT are modified so as to obtain a faster response. The signal is digitized using an analog to digital converter with a sampling frequency of up to 20 MHz. The whole signal produced by the x-ray tube is acquired and analyzed off-line in order to accurately recognize pile-up events and reconstruct the emitted spectrum. The energy resolution has been determined using a calibrated x-ray source. Spectra were validated by comparison of the HVL measured using an ionization chamber.

A massive lesion detection algorithm in mammography

A new algorithm for massive lesion detection in mammography is presented. The algorithm consists in three main steps: 1) reduction of the dimension of the image to be processed through the identification of regions of interest (roi) as candidates for massive lesions; 2) characterization of the RoI by means of suitable feature extraction; 3) pattern classification through supervised neural networks. Suspect regions are detected by searching for local maxima of the pixel grey level intensity. A ring of increasing radius, centered on a maximum, is considered until the mean intensity in the ring decreases to a defined fraction of the maximum. The ROIS thus obtained are described by average, variance, skewness and kurtosis of the intensity distributions at different fractions of the radius. A neural network approach is adopted to classify suspect pathological and healthy pattern. The software has been designed in the framework of the INFN (Istituto Nazionale Fisica Nucleare) research project GPCALMA (Grid Platform for Calma) which recruits physicists and radiologists from different Italian Research Institutions and hospitals to develop software for breast cancer detection.

The MAGIC-5 Project: medical applications on a GRID infrastructure connection

The MAGIC-5 Project aims at developing computer aided detection (CAD) software for medical applications on distributed databases by means of a GRID infrastructure connection. The use of automatic systems for analyzing medical images is of paramount importance in the screening programs, due to the huge amount of data to check. Examples are: mammographies for breast cancer detection, computed-tomography (CT) images for lung cancer analysis, and the positron emission tomography (PET) imaging for the early diagnosis of the Alzheimer disease. The need for acquiring and analyzing data stored in different locations requires a GRID approach of distributed computing system and associated data management. The GRID technologies allow remote image analysis and interactive online diagnosis, with a relevant reduction of the delays actually associated to the screening programs. From this point of view, the MAGIC-5 Collaboration can be seen as a group of distributed users sharing their resources for implementing different virtual organizations (VO), each one aiming at developing screening programs, tele-training, tele-diagnosis and epidemiologic studies for a particular pathology.

The CALMA system: an artificial neural network method for detecting masses and microcalcifications in digitized mammograms

Abstract The Computer Assisted Library for MAmmography (CALMA) project is a 5 years plan developed in a physics research frame in collaboration between Istituto Nazionale di Fisica Nucleare and many Italian hospitals. At present a large database of digitized mammographic images (more than 6000) was collected and a software based on neural network algorithms for the search of suspicious breast lesions was developed. Two tools are available: a microcalcification clusters hunter, based on supervised and unsupervised feedforward neural network, and a massive lesion searcher, based on a hibrid approach. Both the algorithms analyzed preprocessed digitized images by high-frequency filters. Clinical tests were performed to evaluate sensitivity and specificity of the system, considering the system as alone and as second reader. Results show that the system is ready to be implemented by medical industry. The CALMA project, just ended, has its natural development in the Grid Platform for CALMA project, where distributed users join common resources (images, tools, and statistical analysis).

A novel Active Contour Model algorithm for contour detection in complex objects

A new active contour model (ACM) algorithm for the detection of the contour of bi-dimensional regions is presented. The algorithm is based on the simulation of an elastic band glued to the contour of the region under analysis. As a result a local convex hull is obtained, where the radius of the concave regions included by the elastic band is defined by properly tuning a parameter. A dedicated application to medical images is presented. The algorithm is part of a segmentation system able to extract the lung volume from 3D CT scans. The effectiveness of the algorithm is evaluated on a database of 15 low-dose CT scans (about 320 sectional images per CT), including 26 nodules. No pathological structure is missing after the lung volume segmentation, while a reduction of the volume to analyze is obtained to about 15% of the total volume of the original CT scan, and 25% of the chest volume.

GPCALMA, a mammographic CAD in a GRID connection

Abstract The purpose of this work is the development of an automatic system which could be useful for radiologists in the investigation of breast cancer. A breast neoplasia is often marked by the presence of microcalcifications and massive lesions in the mammogram: hence the need for tools able to recognize such lesions at an early stage. Grid Platform Computer-Assisted Library for MAmmography (GPCALMA), a collaboration among Italian physicists and radiologists, has built a large distributed database of digitized mammographic images (at this moment about 5500 images corresponding to 1650 patients). This collaboration has developed a Computer Aided Detection (CAD) system which, installed in an integrated station, can also be used for digitization, as archive, and to perform statistical analysis. With a GRID configuration, it would be possible for clinicians tele- and co-working in new and innovative groupings (‘virtual organizations’), and using the whole database, by the GPCALMA tools, several analysis can be performed. Furthermore, GPCALMA allows to be abreast of the CAD technical progressing into several hospital locations always with remote working by GRID connection. We report in the results obtained by the GPCALMA CAD software implemented with a GRID connection.

A new CAD system for lung nodule detection on low dose CT validated on publicly research database

Purpose Conventional chest X-ray is still the most common radiological imaging to detect pulmonary metastases and nodules. The detection rate of such nodules, however, is limited due to different experiences of the radiologists and depends furthermore critically on the image quality. The role of conventional X-ray as a screening tool is critically discussed [1]. The enhanced and automated detection of pulmonary nodules but also the automated detection of other alterations of the lung including fibrotic and emphysematic structures is, thus of essential importance. In order to achieve a reliable and accurate analysis computer-based image analysis quality might offer additional help. Currently available CAD-solutions will shortly be discussed within this invited lecture with special focus on clinical results and further needs of development. Methods Recently established CAD-solutions offer the opportunity of an automated detection of pulmonary nodules, there is no software-based limitation regarding size or location of the nodules and masses [2,3,4], see Figs. 1, 2, 3. The most common software solutions do not allow an automated size calculation, yet. Application of CAD is not limited to p-aviews but also applicable to portable images. Detection on lateral views are not available in all CAD-systems on the market, yet. Morphological parameters are not automatically quantified, yet. Thus CAD is mainly focussed on the detection of any nodules, for the time being. In order to improve image quality, as an additional tool, a bone and rib-reducing software has been established [5], see Fig. 1. Effect of CAD was analyzed focussing on potential influencing parameters, including histopathology, size, location, imaging quality. Additionally effects of CAD on detection rate and observer reliability were analyzed and typical imaging characteristics were classified as most common reasons for false positive or negative analyses. Finally, the role of currently available CAD in the detection of conventionally occult malignant lesions was evaluated. Personal analyses base on 550 chest X-rays (300 proven malignancies, 100 benign cases including portables, 100 CT-proven unsuspicious cases and 50 cases with missed/occult malignant lesions). Analysis was performed using OnGuard Europe 5.0, Riverain Medical USA including Softview mode. CAD marks (region of interest, ROI, were matched to CT and histopathology in consensus by two experienced radiologists. Sizes of the lesions were scored as S1: up to 10mm, S2: up to 20mm; S3: up to 30mm; S4: up to 40mm; S5: up to 50mm; S6 all other lesions. Results Own results