Multitask Deep Learning Reconstruction and Localization of Lesions in Limited Angle Diffuse Optical Tomography

<div>Diffuse optical tomography (DOT) leverages near-infrared light propagation through tissue to assess its optical properties and identify abnormalities. </div><div>DOT image reconstruction from limited-angle data acquisition is severely ill-posed due to the highly scattered photons in the medium and the relatively small number of collected projections. Reconstructions are thus commonly marred by artifacts and, as a result, it is difficult to obtain accurate reconstruction of target objects, e.g., malignant lesions.</div><div>Reconstruction does not always ensure good localization of small lesions. Furthermore, conventional optimization-based reconstruction methods are computationally expensive, rendering them too slow for real-time imaging applications.</div><div>Our goal is to develop a fast and accurate image reconstruction method using deep learning, where multitask learning ensures accurate lesion localization in addition to improved reconstruction.</div><div>We apply spatial-wise attention and a distance transform based loss function in a novel multitask learning formulation to improve localization and reconstruction compared to single-task optimized methods.</div><div>Given the scarcity of real-world sensor-image pairs required for training supervised deep learning models,</div><div>we leverage physics-based simulation to generate synthetic datasets and use a transfer learning module to align the sensor domain distribution between in silico and real-world data, while taking advantage of cross-domain learning.</div><div>Both quantitative and qualitative results on phantom and real data indicate the superiority of our multitask method in the reconstruction and localization of lesions in tissue compared to state-of-the-art methods.</div><div>The results demonstrate that multitask learning provides sharper and more accurate reconstruction.</div>

Multitask Deep Learning Reconstruction and Localization of Lesions in Limited Angle Diffuse Optical Tomography

<div>Diffuse optical tomography (DOT) leverages near-infrared light propagation through tissue to assess its optical properties and identify abnormalities. </div><div>DOT image reconstruction from limited-angle data acquisition is severely ill-posed due to the highly scattered photons in the medium and the relatively small number of collected projections. Reconstructions are thus commonly marred by artifacts and, as a result, it is difficult to obtain accurate reconstruction of target objects, e.g., malignant lesions.</div><div>Reconstruction does not always ensure good localization of small lesions. Furthermore, conventional optimization-based reconstruction methods are computationally expensive, rendering them too slow for real-time imaging applications.</div><div>Our goal is to develop a fast and accurate image reconstruction method using deep learning, where multitask learning ensures accurate lesion localization in addition to improved reconstruction.</div><div>We apply spatial-wise attention and a distance transform based loss function in a novel multitask learning formulation to improve localization and reconstruction compared to single-task optimized methods.</div><div>Given the scarcity of real-world sensor-image pairs required for training supervised deep learning models,</div><div>we leverage physics-based simulation to generate synthetic datasets and use a transfer learning module to align the sensor domain distribution between in silico and real-world data, while taking advantage of cross-domain learning.</div><div>Both quantitative and qualitative results on phantom and real data indicate the superiority of our multitask method in the reconstruction and localization of lesions in tissue compared to state-of-the-art methods.</div><div>The results demonstrate that multitask learning provides sharper and more accurate reconstruction.</div>

A Modified Helmholtz Equation Least Squares Method for Reconstructing Vibroacoustic Quantities on an Arbitrarily Shaped Vibrating Structure

A modified Helmholtz equation least-square (HELS) method is developed to reconstruct vibroacoustic quantities on an arbitrarily shaped vibrating structure. Unlike the traditional nearfield acoustical holography that relies on the acoustic pressures collected on a hologram surface at a short stand-off distance to a target structure, this modified HELS method takes the partial normal surface velocities and partial acoustic pressures as the input data. The advantages of this approach include but not limited to: (1) The normal surface velocities that represent the nearfield effects are collected directly, which lead to a more accurate reconstruction of the normal surface velocity distribution; (2) The field acoustic pressures are also measured, which leads to a more accurate reconstruction of the acoustic pressure on the source surface as well as in the field; and (3) There is no need to measure the normal surface velocities over the entire surface, which makes this approach quite appealing in practice because most vibrating structures do not allow for measuring the normal surface velocities over the entire source surface as there are always obstacles or constrains around a target structure. Needless to say, regularization is necessary in reconstruction process since all inverse problems are mathematically ill-posed. To validate this approach, both numerical simulations and experimental results are presented. An optimal reconstruction scheme is developed via numerical simulations to achieve the most cost-effective reconstruction results for practical applications.

The Location of the Fibular Tunnel for Anatomically Accurate Reconstruction of the Lateral Ankle Ligament: A Cadaveric Study

We aimed to describe the location of fibular footprint of each anterior talofibular ligament (ATFL) and calcaneofibular ligament (CFL), as well as their common origin in relation to bony landmarks of the fibula in order to determine the location of the fibular tunnel. In 105 ankle specimens, the center of the footprints of the ATFL and CFL (cATFL and cCFL, respectively) and the intersection point of their origin (intATFL-CFL) were investigated, and the distances from selected bony landmarks (the articular tip (AT) and the inferior tip (IT) of the fibula) were measured. Forty-two (40%) specimens had single-bundle ATFL, and 63 (60%) had double-bundle patterns. The distance between intATFL-CFL and IT was 12.0 ± 2.5 mm , and a significant difference was observed between the two groups ( p = 0.001 ). Moreover, the ratio of the intATFL-CFL location based on the anterior fibular border for all cadavers was 0.386. The present study suggests a reference ratio that can help surgeons locate the fibular tunnel for a more anatomically accurate reconstruction of the lateral ankle ligament. Also, it may be necessary to make a difference in the location of the fibular tunnel according to the number of ATFL bundles during surgery.