Deep learning has revolutionized medical image analysis, promising to significantly improve the precision of diagnoses and therapies through advanced segmentation methods. However, the efficacy of deep neural networks is often compromised by the prevalence of imperfect medical labels, while acquiring large-scale, accurately labeled data remains a prohibitive challenge. To address the imperfect label issue, we introduce a novel learning framework that iteratively optimizes both a neural network and its label set to enhance segmentation accuracy. This framework operates through two steps: initially, it robustly trains on a dataset with label noise, distinguishing between clean and noisy labels, and subsequently, it refines noisy labels based on high-confidence predictions from the robust network. By applying this method, not only is the network trained more effectively on imperfect data, but the dataset is progressively cleaned and expanded. Our evaluations are conducted on retina Optical Coherence Tomography datasets using U-Net and SegNet architectures, and demonstrate substantial improvements in segmentation accuracy and data quality, advancing the capabilities of weakly supervised segmentation in medical imaging.
In this project, we propose a deep learning based weakly supervised learning algorithm for cardiac adipose tissue segmentation using image-level labels. Based on ReLayNet, our proposed method can automatically segment the adipose tissue from normal myocardium tissue in pixel level. Compared with fully supervised learning methods, our model achieves competitive segmentation results on both accuracy and Dice coefficient within a database of OCT images of human cardiac tissue. Combined with the OCT image, the predicted adipose map could provide additional information for the guidance of cardiac radio frequency ablation.
Optical coherence tomography (OCT) has emerged as a promising image modality to characterize biological tissues. With axio-lateral resolutions at the micron-level, OCT images provide detailed morphological information and enable applications such as optical biopsy and virtual histology for clinical needs. Image enhancement is typically required for morphological segmentation, to improve boundary localization, rather than enrich detailed tissue information. We propose to formulate image enhancement as an image simplification task such that tissue layers are smoothed while contours are enhanced. For this purpose, we exploit a Total Variation sparsity-based image reconstruction, inspired by the Compressed Sensing (CS) theory, but specialized for images with structures arranged in layers. We demonstrate the potential of our approach on OCT human heart and retinal images for layers segmentation. We also compare our image enhancement capabilities to the state-of-the-art denoising techniques.
In this work, Compressed Sensing (CS) is investigated as a denoising tool in bioimaging. The denoising algorithm exploits multiple CS reconstructions, taking advantage of the robustness of CS in the presence of noise via regularized reconstructions and the properties of the Fourier transform of bioimages. Multiple reconstructions at low sampling rates are combined to generate high quality denoised images using several sparsity constraints. We present different combination methods for the CS reconstructions and quantitatively compare the performance of our denoising methods to state-of-the-art ones.
In this paper, we investigate how the CS framework can be adapted to biological video microscopy acquisition problems. We first consider a frame-by-frame linear acquisition model in the Fourier domain of the signal, and discuss the relevance of several sparsity models that can be used to drive the reconstruction of the whole video sequence. Then, we switch to a non-linear acquisition model – therefore beyond the “pure” CS framework – in which only the modulus of the Fourier transform of the signal is acquired: by exploiting sparsity properties similar to the one used in the linear acquisition case, we demonstrate the feasibility of a phase retrieval reconstruction procedure applied to video signals.
When undergoing reconstruction through a compressed sensing scheme, real microscopic images are affected
by various artifacts that lead to detail loss. Here, we discuss how the sampling strategy and the subsampling
rate affect the compressed sensing reconstruction, and how they should be determined according to a targeted
accuracy level of the reconstruction. We investigate the relevance and limits of theoretical results through
several numerical reconstructions of test images. We discuss the quantitative and qualitative artifacts that affect
the reconstructed signal when reducing the number of measurements in the Fourier domain. We conclude by
extending our results to real microscopic images.
KEYWORDS: Image segmentation, Convolution, Angiography, 3D image processing, Liver, Mathematical modeling, Medical imaging, 3D modeling, Image processing, Image processing algorithms and systems
In the context of mathematical modeling of complex vessel tree structures with deformable models, we present
a novel level set formulation to evolve both the vessel surface and its centerline. The implicit function is
computed as the convolution of a geometric primitive, representing the centerline, with localized kernels of
continuously-varying scales allowing accurate estimation of the vessel width. The centerline itself is derived
as the characteristic function of an underlying signed medialness function, to enforce a tubular shape for the
segmented object, and evolves under shape and medialness constraints. Given a set of initial medial loci and
radii, this representation first allows for simultaneous recovery of the vessels centerlines and radii, thus enabling
surface reconstruction. Secondly, due to the topological adaptivity of the level set segmentation setting, it can
handle tree-like structures and bifurcations without additional junction detection schemes nor user inputs. We
discuss the shape parameters involved, their tuning and their influence on the control of the segmented shapes,
and we present some segmentation results on synthetic images, 2D angiographies, 3D rotational angiographies
and 3D-CT scans.
KEYWORDS: Signal to noise ratio, Denoising, Microscopy, Luminescence, Interference (communication), Image restoration, Image acquisition, Compressed sensing, Molecules, Process control
Noise level and photobleaching are cross-dependent problems in biological fluorescence microscopy. Indeed,
observation of fluorescent molecules is challenged by photobleaching, a phenomenon whereby the fluorophores
are degraded by the excitation light. One way to control this process is by reducing the intensity of the light or the
time exposure, but it comes at the price of decreasing the signal-to-noise ratio (SNR). Although a host of denoising
methods have been developed to increase the SNR, most are post-processing techniques and require full data
acquisition. In this paper we propose a novel technique, based on Compressed Sensing (CS) that simultaneously
enables reduction of exposure time or excitation light level and improvement of image SNR. Our CS-based
method can simultaneously acquire and denoise data, based on statistical properties of the CS optimality, noise
reconstruction characteristics and signal modeling applied to microscopy images with low SNR. The proposed
approach is an experimental optimization combining sequential CS reconstructions in a multiscale framework
to perform image denoising. Simulated and practical experiments on fluorescence image data demonstrate that
thanks to CS denoising we obtain images with similar or increased SNR while still being able to reduce exposure
times. Such results open the gate to new mathematical imaging protocols, offering the opportunity to reduce
photobleaching and help biological applications based on fluorescence microscopy.
Standard spatial compounding, via averaging acquisitions from different angles, has proved to be an efficient technique for speckle pattern reduction in ultrasound B-mode images. However, the resulting images may be blurred due to the averaging of point spread functions and the misalignment of the different views. These blurring artefacts result in a loss of important anatomical features that may be critical for medical diagnosis. In this paper, we evaluate some spatial compounding techniques, focusing on how to combine the different acquisitions. The evaluated methods are: weighed averaging, wavelet coefficient fusion and multiview deconvolution. To some extent, these techniques take into account the limitations of spatial compounding, by proposing alternative fusion methods that can reduce speckle artefacts while preserving standard spatial resolution and anatomical features.
We experimented these compounding methods with synthetic images to show that these advanced techniques could outperform traditional averaging. In particular, multiview deconvolution techniques performed best, showing improvement in respect to averaging (6.81 dB) for realistic levels of speckle noise and spatial degradation. Wavelet fusion technique ranked second (2.25 dB), and weighted average third (0.70 dB). On the other hand, weighted averaging was the least time consuming, followed by wavelet fusion (x2) and multiview deconvolution (x5). Wavelet fusion offered an interesting trade-off between performance and computational cost.
Experiments on 3D breast ultrasound imaging, showed consistent results with those obtained on synthetic images. Tissue was linearly scanned with a 2D probe in different directions, and volumes were compounded using the aforementioned techniques. This resulted in a high-resolution volume, with better tissue delineation and less speckle patterning.
KEYWORDS: Data modeling, 3D modeling, Image segmentation, Motion models, Echocardiography, Ultrasonography, Heart, 3D image processing, Spherical lenses, 3D scanning
This paper presents a new formulation of a deformable model segmentation in prolate spheroidal coordinates for segmentation of 3D cardiac echocardiography data. The prolate spheroidal coordinate system enables a representation of the segmented surface with descriptors specifically adapted to the "ellipsoidal" shape of the ventricle. A simple data energy term, based on gray-level information, guides the segmentation. The segmentation framework provides a very fast and simple algorithm to evolve an initial ellipsoidal object towards the endocardial surface of the myocardium with near real-time deformations. With near real-time performance, additional constraints on landmark points, can be used interactively to prevent leakage of the surface.
Micro-array protein crystal images are now routinely acquired automatically by CCD cameras. High-throughput automatic classification of protein crystals requires to alleviation of the time-consuming task of manual visual inspection. We propose a classification framework combined with a multi-scale image processing method for recognizing protein crystals and precipitates versus clear drops. The main two points of the processing method are the multi-scale Laplacian pyramid filters and histogram analysis techniques to find an effective feature vector. The processing steps include: 1. Tray well cropping using Radon Transform; 2. Droplet cropping using an ellipsoid Hough Transform; 3. Multi-scale image separation with Laplacian pyramidal filters; 4. Feature vector extraction from the histogram of the multi-scale boundary images. The feature vector combines geometric and texture features of each image and provides input to a feed forward binomial neural network classifier. Using human (expert crystallographers) classified images as ground truth, the current experimental results gave 86% true positive and 94% true negative rates (average true percentage is 90%) using an image database which contained over 2,000 images. To enable NESG collaborators to carry our crystal classification, a web-based Matlab server was also developed. Users at other locations on the internet can input micro-array crystal image folders and parameters for training and testing processes through a friendly web interface. Recognition results are shown on the client side website and may be downloaded by a remote user as an Excel spreadsheet file.
With relatively high frame rates and the ability to acquire volume data sets with a stationary transducer, 3D ultrasound systems, based on matrix phased array transducers, provide valuable three-dimensional information, from which quantitative measures of cardiac function can be extracted. Such analyses require segmentation and visual tracking of the left ventricular endocardial border. Due to the large size of the volumetric data sets, manual tracing of the endocardial border is tedious and impractical for clinical applications. Therefore the development of automatic methods for tracking three-dimensional endocardial motion is essential. In this study, we evaluate a four-dimensional optical flow motion tracking algorithm to determine its capability to follow the endocardial border in three dimensional ultrasound data through time. The four-dimensional optical flow method was implemented using three-dimensional correlation. We tested the algorithm on an experimental open-chest dog data set and a clinical data set acquired with a Philips' iE33 three-dimensional ultrasound machine. Initialized with left ventricular endocardial data points obtained from manual tracing at end-diastole, the algorithm automatically tracked these points frame by frame through the whole cardiac cycle.
A finite element surface was fitted through the data points obtained by both optical flow tracking and manual tracing by an experienced observer for quantitative comparison of the results. Parameterization of the finite element surfaces was performed and maps displaying relative differences between the manual and semi-automatic methods were compared.
The results showed good consistency between manual tracing and optical flow estimation on 73% of the entire surface with fewer than 10% difference. In addition, the optical flow motion tracking algorithm greatly reduced processing time (about 94% reduction compared to human involvement per cardiac cycle) for analyzing cardiac function in three-dimensional ultrasound data sets.
KEYWORDS: Image segmentation, Magnetic resonance imaging, Brain, Tissues, 3D modeling, 3D image processing, Data modeling, Binary data, Neuroimaging, Clinical research
Segmentation of three-dimensional anatomical brain images into tissue classes has applications in both clinical and research settings. This paper presents the implementation and quantitative evaluation of a four-phase three-dimensional active contour implemented with a level set framework for automated segmentation of brain MRIs. The segmentation algorithm performs an optimal partitioning of three-dimensional data based on homogeneity measures that naturally evolves to the extraction of different tissue types in the brain. Random seed initialization was used to speed up numerical computation and avoid the need for a priori information. This random initialization ensures robustness of the method to variation of user expertise, biased a priori information and errors in input information that could be influenced by variations in image quality. Experimentation on three MRI brain data sets showed that an optimal partitioning successfully labeled regions that accurately identified white matter, gray matter and cerebrospinal fluid in the ventricles. Quantitative evaluation of the segmentation was performed with comparison to manually labeled data and computed false positive and false negative assignments of voxels for the three organs. We report high accuracy for the two comparison cases. These results demonstrate the efficiency and flexibility of this segmentation framework to perform the challenging task of automatically extracting brain tissue volume contours.
Three-dimensional ultrasound machines based on matrix phased-array transducers are gaining predominance for real-time dynamic screening in cardiac and obstetric practice. These transducers array acquire three-dimensional data in spherical coordinates along lines tiled in azimuth and elevation angles at incremental depth. This study aims at evaluating fast filtering and scan conversion algorithms applied in the spherical domain prior to visualization into Cartesian coordinates for visual quality and spatial measurement accuracy.
Fast 3d scan conversion algorithms were implemented and with different order interpolation kernels. Downsizing and smoothing of sampling artifacts were integrated in the scan conversion process. In addition, a denoising scheme for spherical coordinate data with 3d anisotropic diffusion was implemented and applied prior to scan conversion to improve image quality. Reconstruction results under different parameter settings, such as different interpolation kernels, scaling factor, smoothing options, and denoising, are reported. Image quality was evaluated on several data sets via visual inspections and measurements of cylinder objects dimensions. Error measurements of the cylinder's radius, reported in this paper, show that the proposed fast scan conversion algorithm can correctly reconstruct three-dimensional ultrasound in Cartesian coordinates under tuned parameter settings. Denoising via three-dimensional anisotropic diffusion was able to greatly improve the quality of resampled data without affecting the accuracy of spatial information after the modification of the introduction of a variable gradient threshold parameter.
High-resolution multichannel textures are difficult to characterize with simple statistics and the high level of detail makes the selection of a particular contour using classical gradient-based methods not effective. We have developed a hybrid method that combines fuzzy connectedness and Voronoi diagram classification for the segmentation of color and multichannel objects. The multi-step classification process relies on homogeneity measures derived from moment statistics and histogram information. These color features have been optimized to best combine individual channel information in the classification process. The segmentation initialization requires only a set of interior and exterior seed points, minimizing user intervention and the influence of the initialization on the overall quality of the results. The method was tested on volumes from the Visible Human and on brain multi-protocol MRI data sets. The hybrid segmentation produced robust, rapid and finely detailed contours with good visual accuracy. The addition of quantized statistics and color histogram distances as classification features improved the robustness of the method with regards to initialization when compared to our original implementation.
Speckle noise corrupts ultrasonic data by introducing sharp changes in an echocardiographic image intensity profile, while attenuation alters the intensity of equally significant cardiac structures. These properties introduce inhomogeneity in the spatial domain and suggests that measures based on phase information rather than intensity are more appropriate for denoising and cardiac border detection. The present analysis method relies on the expansion of temporal ultrasonic volume data on complex exponential wavelet-like basis functions called Brushlets. These basis functions decompose a signal into distinct patterns of oriented textures. Projected coefficients are associated with distinct 'brush strokes' of a particular size and orientation. 4D overcomplete brushlet analysis is applied to temporal echocardiographic values. We show that adding the time dimension in the analysis dramatically improves the quality and robustness of the method without adding complexity in the design of a segmentation tool. We have investigated mathematical and empirical methods for identifying the most 'efficient' brush stroke sizes and orientations for decomposition and reconstruction on both phantom and clinical data. In order to determine the 'best tiling' or equivalently, the 'best brushlet basis', we use an entorpy-based information cost metric function. Quantitative validation and clinical applications of this new spatio-temporal analysis tool are reported for balloon phantoms and clinical data sets.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.