Search

In this work, we propose accurate binary Depthwise Separable Convolutional Neural Networks (DSCNNs) for medical image segmentation. The networks are binarized by learning the distribution of weights and activations, and by using parameter-free skip connections in their encoder and decoder structure. We design full precision DSCNNs based on a symmetric encoder-decoder, feature pyramid network with an asymmetric decoder, and spatial pyramid pooling with atrous convolutions strategies for image segmentation. The DSCNNs have 14 X and 8 X fewer number of model parameters and operations, respectively, than standard segmentation networks. The trained full precision DSCNNs are used as baselines to achieve accurate binary DSCNNs. The networks are trained on two medical ultrasound datasets, a public fetal skull dataset and a privileged bladder dataset. The accuracy of the binary DSCNNs are within a 3% drop from the full precision networks on both the medical datasets.

Bone stress injuries (BSI) are caused by repetitive loading patterns that do not allow adequate time for bone remodeling. These injuries are most prevalent among endurance athletes and military personnel, particularly military recruits. Treatment for BSI usually includes some level of activity modification and, with prescribed recovery times of weeks to months, these injuries impose significant time lost from training or duties. The gold standard method for diagnosing and grading stress injuries is MRI. Ultrasound would be an attractive alternative or supplement for BSI diagnosis and recovery monitoring, since it can be applied at point-of-care locations and repeat scans are safe and relatively low cost. We describe the Image Registered Quantitative Ultrasound (IR-QUS) technique, aimed at evaluating BSI on the basis of surface roughness as an indicator of rapid bone remodeling. IR-QUS collects angle-diverse acoustic measurements on a focal point and detects relative increases in diffusive versus specular scattering. We demonstrate the sensitivity of this method to characterize surface roughness on the scale of tens of microns, and show the feasibility of detecting a localized field of fractures on an ex vivo intact bone sample. Experimental results suggest reliable detection of roughened surfaces with a diffuse-to-specular energy metric that increases by 10 dB relative to smooth regions.

MIT Lincoln Laboratory, the Medical Device Realization Center (MEDRC) at MIT, and the Massachusetts General Hospital (MGH) are collaboratively developing a novel optical system that acquires ultrasound images within the human body without physical contact to the patient. The system is termed, non-contact laser ultrasound (N-CLUS) and yields anatomical images in tissue and bone and can also measure elastographic properties, in-vivo, all from an operational standoff of a few inches to several meters as desired. N-CLUS employs a pulsed laser that converts optical energy into ultrasonic waves at the skin surface via photoacoustic mechanisms, while, a laser Doppler vibrometer measures reflected-emerging ultrasonic waves from tissue at depth at the skin surface. The key of the N-CLUS approach is driven by shallow optical absorptivity that creates an acoustic source that enables ultrasound propagation deeper into the tissue.We discuss the motivation of the non-contact laser concept, its development path involving signal generation, skin and eye safe laser measurement, and system design perspectives. Elastogrphic measurements are then demonstrated with determination of bone elastic moduli for beef rib within tissue. N-CLUS images from soft tissue specimens are also compared with commercial ultrasound, showing that the noncontact optical approach may have potential as a viable method in medical ultrasound.

This paper introduces a non-invasive, contrast-enhanced ultrasound (CEUS) infusion method to quantify the health of viable donor livers. The method uses the infusion of microbubbles and their destruction and subsequent replenishment to measure the perfusion rate in the liver microvasculature. The proposed method improves on the previous parameter extraction approaches applied to the flash-replenishment technique by addressing the effects of the microbubble mixing within the perfusate bath and destruction rate. By doing so, the tissue perfusion rate can be extracted from the data even though the microbubble concentration is not constant throughout image acquisition. The measured changes in the tissue perfusion rate showed that CEUS infusion is a viable biomarker for assessing liver health.

The potential of a fully noncontact, standoff, laserultrasound system that acquires ultrasonic images within biological tissue is examined. A pulsed laser converts optical energy into ultrasound via photoacoustic mechanisms, while laser Doppler vibrometry measures emerging ultrasonic waves at the tissue surface. Differing from photoacoustic tomography (PAT), which maps spatial variations in tissue-optical absorptivity in the acoustic near field, the laser ultrasound (LUS) approach developed here, is driven by shallow, non-varying optical absorptivity that creates a laterally consistent acoustic source enabling ultrasound propagation well into the far field. LUS acoustic wave generation is explored in tissue and bone including longitudinal, shear, and Rayleigh wave components. Using information from LUS wave types can yield 1) tissue and bone anatomical images and 2) mechanical property distributions that apply to the emerging field of medical elastography. Imaging capabilities using a demonstration LUS system are also presented for complex bio-tissues. Ultrasonic images compare well with ground truth geometries, orientation, and depth of staged samples. 2D cross-sectional echo reflection images are generated for a phantom limb containing muscle and bone materials and use data inversion techniques to yield the elastic moduli distributions in the specimen.

Using ultrasonic shear wave speed (SWS) estimates has become popular to noninvasively evaluate liver fibrosis, but significant inter-system variability in liver SWS measurements can preclude meaningful comparison of measurements performed with different systems. The RSNA Quantitative Imaging Biomarker Alliance (QIBA) ultrasound SWS committee has been developing elastic and viscoelastic (VE) phantoms to evaluate system dependencies of SWS estimates. The objective of this study is to compare SWS measurements between commercially-available systems using phantoms that have viscoelastic properties similar to those observed in normal and fibrotic liver. CIRS, Inc. fabricated three phantoms using a proprietary oil-water emulsion infused in a Zerdine® hydrogel that were matched in viscoelastic behavior to healthy and fibrotic human liver data. Phantoms were measured at academic, clinical, government and vendor sites using different systems with curvilinear arrays at multiple focal depths (3.0, 4.5 & 7.0 cm). The results of this study show that current-generation ultrasound SWS measurement systems are able to differentiate viscoelastic materials that span healthy to fibrotic liver. The deepest focal depth (7.0 cm) yielded the greatest inter-system variability for each phantom (maximum of 17.7%) as evaluated by IQR. Inter-system variability was consistent across all 3 phantoms and was not a function of stiffness. Median SWS estimates for the greatest outlier system for each phantom/focal depth combination ranged from 12.7–17.6%. Future efforts will include performing more robust statistical analyses of these data, comparing these phantom data trends with viscoelastic digital phantom data, providing vendors with study site data to refine their systems to have more consistent measurements, and integrating these data into the QIBA ultrasound shear wave speed measurement profile.

An interlaboratory study of shear wave speed (SWS) estimation was performed. Commercial shear wave elastography systems from Fibroscan, Philips, Siemens and Supersonic Imagine, as well as several custom laboratory systems, were involved. Fifteen sites were included in the study. CIRS manufactured and donated 11 pairs of custom phantoms designed for the purposes of this investigation. Dynamic mechanical tests of equivalent phantom materials were also performed. The results of this study demonstrate that there is very good agreement among SWS estimation systems, but there are several sources of bias and variance that can be addressed to improve consistency of measurement results.