Francesca Galluzzo

A GPU level-set segmentation framework for 3D Echocardiography

Real-time 3D Echocardiography (RT3DE), providing truly volumetric images of the heart, is a promising imaging modality for cardiac morphology and function assessment. Fast and automatic segmentation of the left ventricle (LV) is essential to enable an efficient quantitative analysis of these 3D data. Recently, we proposed a GPU implementation of the Level-Set (LS) Sparse Field algorithm for 3D image segmentation which combines high computational efficiency and flexibility in the interface evolution. In this work, we make our GPU LS solver able to deal with strongly inhomogeneous images such as the myocardial wall in RT3DE. The applicability of our method in clinical environment was evaluated by measuring LV volumetric parameters on 23 RT3DE exams, and comparing them with reference values from manual contouring. Results show the effectiveness of our framework in performing accurate LV myocardium segmentation in RT3DE near real-time.

A rigorous and efficient GPU implementation of level-set sparse field algorithm

Level-set methods have proven to be powerful and flexible tools in computer vision and medical imaging. Unfortunately, the flexibility of such models has historically resulted in long computational times and therefore limited clinical utility. In this context, we propose the first rigorous GPU implementation of the sparse field algorithm. We show that this model is able to reach high computational efficiency with no reduction in segmentation accuracy compared to its sequential counter-part.

Design of anthropomorphic atherosclerotic carotid artery flow phantoms for ultrasound images

Carotid artery phantoms (CaPs) can be used as test objects to explore novel ways of enhancing the ultrasound based carotid atherosclerosis diagnosis. To achieve this goal CaPs should be anatomically realistic both in terms of geometry, acoustic and physical properties, and should allow to reproduce different pathological conditions. We propose a framework for designing CaPs of healthy and diseased arteries. To verify the framework effectiveness we constructed three CaPs: healthy, with a hard/soft plaque causing a 30%/65% vessel narrowing. Then we acquired CaPs B-mode images and performed their geometric characterization and echogenicity analysis demonstrating the framework effectiveness at realizing anthropomorphic CaPs at low cost, easily reproducing different atherosclerotic conditions.

Improving prostate biopsy protocol with a computer aided detection tool based on semi-supervised learning

Prostate cancer is one of the most frequently diagnosed neoplasy and its presence can only be confirmed by biopsy. Due to the high number of false positives, Computer Aided Detection (CAD) systems can be used to reduce the number of cores requested for an accurate diagnosis. This work proposes a CAD procedure for cancer detection in Ultrasound images based on a learning scheme which exploits a novel semi-supervised learning (SSL) algorithm for reducing data collection effort and avoiding collected data wasting. The ground truth database comprises the RFsignals acquired during biopsies and the corresponding tissue samples histopathological outcome. A comparison to a state-of-art CAD scheme based on supervised learning demonstrates the effectiveness of the proposed SSL procedure at enhancing CAD performance. Experiments on ground truth images from biopsy findings show that the proposed CAD scheme is effective at improving the efficiency of the biopsy protocol.

rtCAB: A tool for reducing unnecessary prostate biopsy cores

With more than 110.000 new cases/year in Europe, prostate cancer (PCa) is one of the most frequent neoplasy. When suspects arise from standard diagnostic methods (i.e. Digital Rectal Exam, Transrectal Ultrasonography (TRUS), PSA level) a prostate biopsy (PBx) is mandatory. As patient discomfort and adverse event probability both grows with core number, it is desirable to reduce the number of PBx cores without negative impinging on diagnose accuracy. The work describes an innovative processing technique called real-time Computer Aided Biopsy (rtCAB) which enhances TRUS video stream with a false color overlay image, and suggests the physician where to sample thus reducing the total number of cores. Our proposal consists in a real-time non-linear classifier which processes the output of an original Maximum Likelihood estimator of Nakagami parameters based on Pade´ Approximant. The resulting algorithm, implemented making full use of CUDA parallel processing capabilities, is capable to deliver frame rates as high as 30 fps. Classification model was trained on a prostate gland adenocarcinoma database (400 PBx cores, 8000 ROIs). Ground truth for each core was established by an expert physician, providing tissue description and illness percentage for each core. The system was tuned for reducing the number of false positives while preserving an acceptable number of false negatives. Comparing to a classical double sextant PBx, the positive prediction value (PPV) of our method is 65% better, with an overall sensitivity of 100%.

A GPU 3D Segmentation Framework for Medical Imaging

In this work we propose a fast and flexible GPU 3D level-set segmentation framework able to handle different segmentation tasks. Experiments on simulated and real images demonstrate the method ability at achieving high computational efficiency with no reduction in segmentation accuracy compared to its sequential counterpart. The method clinical applicability is demonstrated by addressing the task of Left-Ventricle myocardium segmentation in Real-Time 3D Echocardiography.

A fully automated method for carotid plaques segmentation in ultrasound images based on motion estimation and level-set

Ultrasound (US) guided carotid atherosclerosis diagnosis is based on the evaluation of the stenosis degree due to the presence of carotid plaques (CPs) and on CPs composition study. Accurate and automated CPs segmentation is essential to enable these evaluations. In this work, we present a fully automated CPs segmentation method based on level-set with an innovative initialization procedure that makes the approach completely user-independent by exploiting the carotid walls motion analysis. Performance were tested on 10 US image sequences of carotid artery and compared with manual contouring from an expert physician. Results show the effectiveness of our method at performing accurate CPs segmentation in US images without requiring any user intervention.

Learning Features from Medical Radiofrequency Ultrasonic Signals by Independent Component Analysis

This paper proposes the use of independent component analysis (ICA) method for learning features from radio frequency (RF) ultrasonic signals. Conventional feature extractors usually suffer from limitations caused by some of their assumptions about the structure of imaged organ and the interaction between ultrasonic signal and tissue. ICA, on the other hand, is a data-driven approach which learns efficient representation of data by maximizing independence of some basis functions that describe the important structures of the data. It has less restrictive considerations about formation of data and as a consequence, could adapt itself to the characteristics of data with different natures. These features of ICA make it a proper candidate for dealing with the problem of feature extraction from medical ultrasound signals where variations in tissue structures and data acquisition conditions are considerable. Experimental results on raw backscattered RF signals of prostate gland show favorable performance of ICA as compared with the conventional methods designed for extracting feature from RF signals.

Soft-label reinforced rtCAB for guided prostate tissue sampling

In this paper a real-time computer-aided biopsy (rtCAB) system is presented to support prostate cancer diagnosis. Different types of features are extracted from trans-rectal ultrasound data and an ensemble learning algorithm is used in classification phase. A new label assignment method is also employed to provide soft or crisp class labels for uncertain data. The proposed model could be implemented in parallel on GPU using CUDA platform to provide real-time support to physician during biopsy. Experiments on ground truth images from biopsy finding demonstrate that the proposed approach can properly deal with uncertain data and is able to provide better results than some examined supervised and semi-supervised classifiers.