Orthogonal Images Research Articles

Three-dimensional Visualization in Diagnostic Imaging: A Renaissance in Radiology.

Supplemental material is available for this article.

Supervised vehicle trajectory prediction using orthogonal image map for urban automated driving

Supervised vehicle trajectory prediction using orthogonal image map for urban automated driving

Real-time motion monitoring using orthogonal cine MRI during MR-guided adaptive radiation therapy for abdominal tumors on 1.5T MR-Linac.

Real-time motion monitoring (RTMM) is necessary for accurate motion management of intrafraction motions during radiation therapy (RT). Building upon a previous study, this work develops and tests an improved RTMM technique based on real-time orthogonal cine magnetic resonance imaging (MRI) acquired during magnetic resonance-guided adaptive RT (MRgART) for abdominal tumors on MR-Linac. A motion monitoring research package (MMRP) was developed and tested for RTMM based on template rigid registration between beam-on real-time orthogonal cine MRI and pre-beam daily reference 3D-MRI (baseline). The MRI data acquired under free-breathing during the routine MRgART on a 1.5T MR-Linac for 18 patients with abdominal malignancies of 8 liver, 4 adrenal glands (renal fossa) and 6 pancreas cases were used to evaluate the MMRP package. For each patient, a 3D mid-position image derived from an in-house daily 4D-MRI was used to define a target mask or a surrogate sub-region encompassing the target. Additionally, an exploratory case reviewed for an MRI dataset of a healthy volunteer acquired under both free-breathing and deep inspiration breath-hold (DIBH) was used to test how effectively the RTMM using the MMRP can address through-plane motion (TPM). For all cases, the 2D T2/T1-weighted cine MRIs were captured with a temporal resolution of 200 ms interleaved between coronal and sagittal orientations. Manually delineated contours on the cine frames were used as the ground-truth motion. Common visible vessels and segments of target boundaries in proximity to the target were used as anatomical landmarks for reproducible delineations on both the 3D and the cine MRI images. Standard deviation of the error (SDE) between the ground-truth and the measured target motion from the MMRP package were analyzed to evaluate the RTMM accuracy. The maximum target motion (MTM) was measured on the 4D-MRI for all cases during free-breathing. The mean (range) centroid motions for the 13 abdominal tumor cases were 7.69 (4.71-11.15), 1.73 (0.81-3.05), and 2.71 (1.45-3.93) mm with an overall accuracy of <2 mm in the superior-inferior (SI), the left-right (LR), and the anterior-posterior (AP) directions, respectively. The mean (range) of the MTM from the 4D-MRI was 7.38 (2-11) mm in the SI direction, smaller than the monitored motion of centroid, demonstrating the importance of the real-time motion capture. For the remaining patient cases, the ground-truth delineation was challenging under free-breathing due to the target deformation and the large TPM in the AP direction, the implant-induced image artifacts, and/or the suboptimal image plane selection. These cases were evaluated based on visual assessment. For the healthy volunteer, the TPM of the target was significant under free-breathing which degraded the RTMM accuracy. RTMM accuracy of <2 mm was achieved under DIBH, indicating DIBH is an effective method to address large TPM. We have successfully developed and tested the use of a template-based registration method for an accurate RTMM of abdominal targets during MRgART on a 1.5T MR-Linac without using injected contrast agents or radio-opaque implants. DIBH may be used to effectively reduce or eliminate TPM of abdominal targets during RTMM. This article is protected by copyright. All rights reserved.

2D/3D Non-Rigid Image Registration via Two Orthogonal X-ray Projection Images for Lung Tumor Tracking

Two-dimensional (2D)/three-dimensional (3D) registration is critical in clinical applications. However, existing methods suffer from long alignment times and high doses. In this paper, a non-rigid 2D/3D registration method based on deep learning with orthogonal angle projections is proposed. The application can quickly achieve alignment using only two orthogonal angle projections. We tested the method with lungs (with and without tumors) and phantom data. The results show that the Dice and normalized cross-correlations are greater than 0.97 and 0.92, respectively, and the registration time is less than 1.2 seconds. In addition, the proposed model showed the ability to track lung tumors, highlighting the clinical potential of the proposed method.

Interrelation of Symmetry and Antisymmetry of Quasi-Orthogonal Cyclic Matrices with Prime Numbers

Quasi-orthogonal Hadamard matrices and Mersenne matrices with two and three values of the elements, used in digital data processing, are considered, as well as the basis of error-correcting codes and algorithms for transforming orthogonal images. Attention is paid to the structures of cyclic matrices with symmetries and antisymmetries. The connection between symmetry and antisymmetry of structures of cyclic Hadamard and Mersenne matrices on a orders equal to prime numbers, products of close primes, composite numbers, powers of a prime number is shown. Separately, orders equal to the degrees of the prime number 2 are distinguished, both the orders of Hadamard matrices and the basis of the composite orders of Mersenne matrices of block structures with two element values. It is shown that symmetric Hadamard matrices of cyclic and bicyclic structures, according to the extended Riser boundary, do not exist on orders above 32. Mersenne matrices of composite orders belonging to the sequence of Mersenne numbers 2k ‒ 1 nested in the sequence of orders of the main family of Mersenne matrices 4t ‒ 1 exist in a symmetric and antisymmetric form. For orders equal to the powers of a prime number, Mersenne matrices exist in the form of block-diagonal constructions with three element values. The value of prime power determines the number of blocks along the diagonal of the matrix on which the elements with the third value are located. The cyclic blocks are symmetrical and antisymmetric.

A universal orthogonal imaging platform for living-cell RNA detection using fluorogenic RNA aptamers

We report a dual-color orthogonal miLS imaging platform for sequence-specific miRNA imaging in different cells.

3DSRNet: 3-D Spine Reconstruction Network Using 2-D Orthogonal X-Ray Images Based on Deep Learning

3DSRNet: 3-D Spine Reconstruction Network Using 2-D Orthogonal X-Ray Images Based on Deep Learning

High-Resolution 3D MRI With Deep Generative Networks via Novel Slice-Profile Transformation Super-Resolution.

High-resolution magnetic resonance imaging (MRI) sequences, such as 3D turbo or fast spin-echo (TSE/FSE) imaging, are clinically desirable but suffer from long scanning time-related blurring when reformatted into preferred orientations. Instead, multi-slice two-dimensional (2D) TSE imaging is commonly used because of its high in-plane resolution but is limited clinically by poor through-plane resolution due to elongated voxels and the inability to generate multi-planar reformations due to staircase artifacts. Therefore, multiple 2D TSE scans are acquired in various orthogonal imaging planes, increasing the overall MRI scan time. In this study, we propose a novel slice-profile transformation super-resolution (SPTSR) framework with deep generative learning for through-plane super-resolution (SR) of multi-slice 2D TSE imaging. The deep generative networks were trained by synthesized low-resolution training input via slice-profile downsampling (SP-DS), and the trained networks inferred on the slice profile convolved (SP-conv) testing input for 5.5x through-plane SR. The network output was further slice-profile deconvolved (SP-deconv) to achieve an isotropic super-resolution. Compared to SMORE SR method and the networks trained by conventional downsampling, our SPTSR framework demonstrated the best overall image quality from 50 testing cases, evaluated by two abdominal radiologists. The quantitative analysis cross-validated the expert reader study results. 3D simulation experiments confirmed the quantitative improvement of the proposed SPTSR and the effectiveness of the SP-deconv step, compared to 3D ground-truths. Ablation studies were conducted on the individual contributions of SP-DS and SP-conv, networks structure, training dataset size, and different slice profiles.

Feasibility study of deep learning-based markerless real-time lung tumor tracking with orthogonal X-ray projection images.

The feasibility of a deep learning-based markerless real-time tumor tracking (RTTT) method was retrospectively studied with orthogonal kV X-ray images and clinical tracking records acquired during lung cancer treatment. Ten patients with lung cancer treated with marker-implanted RTTT were included. The prescription dose was 50Gy in four fractions, using seven- to nine-port non-coplanar static beams. This corresponds to 14-18 X-ray tube angles for an orthogonal X-ray imaging system rotating with the gantry. All patients underwent 10 respiratory phases four-dimensional computed tomography. After a data augmentation approach, for each X-ray tube angle of a patient, 2250 digitally reconstructed radiograph (DRR) images with gross tumor volume (GTV) contour labeled were obtained. These images were adopted to train the patient and X-ray tube angle-specific GTV contour prediction model. During the testing, the model trained with DRR images predicted GTV contour on X-ray projection images acquired during treatment. The predicted three-dimensional (3D) positions of the GTV were calculated based on the centroids of the contours in the orthogonal images. The 3D positions of GTV determined by the marker-implanted RTTT during the treatment were considered as the ground truth. The 3D deviations between the prediction and the ground truth were calculated to evaluate the performance of the model. The median GTV volume and motion range were 7.42 (range, 1.18-25.74) cm3 and 22 (range, 11-28) mm, respectively. In total, 8993 3D position comparisons were included. The mean calculation time was 85ms per image. The overall median value of the 3D deviation was 2.27 (interquartile range: 1.66-2.95) mm. The probability of the 3D deviation smaller than 5mm was 93.6%. The evaluation results and calculation efficiency show the proposed deep learning-based markerless RTTT method may be feasible for patients with lung cancer.

Orthogonal dual-channel single-camera spectral imaging for online measurement of three-dimensional space–time evolution of temperature field in laser melt pool

Current temperature measurement technology in laser manufacturing cannot obtain the internal three-dimensional (3D) temperature distribution of the melt pool. Necessary information for laser-manufacturing process optimization is lacking, including the melt pool depth and depth-to-width ratio. We propose a method for 3D temperature measurement based on orthogonal dual-channel single-camera spectral imaging. The 3D temperature field distribution of the melt pool can be obtained by collecting projected radiation data from two perpendicular angles. Use of a single monochrome camera for spectral imaging at three wavelengths ensures excellent spatial and temporal consistency during measurements. To verify the feasibility of the method, we measured the 3D temperature field in a melt pool of 304 stainless steel irradiated by a Gaussian laser beam (continuous laser, 500 W) with different irradiation times. The results indicate that the temperature field exhibits an approximately axisymmetric distribution on each cross-section at different depths. The temperature is highest in the center and gradually decreases around it. As the depth increases, the high-temperature area (>1730 K) gradually becomes smaller, and the temperature of each cross-section decreases. The maximum temperature, depth, width, and depth-to-width ratio of the melt pool at different times were obtained from the measured 3D temperature distributions. With an irradiation time of 0–80 s, the maximum temperature of the melt pool exhibited three stages: rapid increase (20–2230 K), slow increase (2230–2280 K), and stabilization. The growth of the width and depth of the melt pool indicated a similar process, increasing rapidly at first and gradually slowing. The depth-to-width ratio decreased rapidly at first and gradually stabilized. The proposed method for 3D temperature measurement has potential use in laser manufacturing for process control and quality improvement.

Use of 3D planning of HDR brachytherapy in patients with cervical cancer

Background. Combined radiation therapy is the «gold standard» for the treatment of stages IIB-IVA of cervical cancer (cervical cancer). Given the data of statistical indicators of cervical cancer, the study of new approaches to treatment and implementation in clinical practice of modern technologies of radiation therapy of widespread cervical cancer is an extremely important area.&#x0D; Purpose – evaluate the benefits of using CT topometric preparation in planning brachytherapy for cervical cancer.&#x0D; Materials and Methods. 3D planning of brachytherapy of 24 patients with locally advanced cervical cancer (FIGO IIIB stage), who underwent a radical program of combined radiation therapy (remote radiation therapy with chemical modification with cisplatin followed by HDR-brachytherapy (High dose rate brachytherapy)).&#x0D; Results. The paper presents the results of the analysis of differences in brachytherapy planning according to orthogonal images and CT images. The analysis of the influence of the constitutional features of patients revealed that there are differences in the topographic anatomy of the pelvis depending on the surface area of the body. In patients with a body surface area of up to 1,65 m2 , a more intimate fit of the intestine to the target volume of irradiation was noted. No similar planning difficulties were found in patients with a body surface area above 1,65 m2 .&#x0D; Conclusions. The use of 3D images in the planning of intracavitary irradiation in patients with cervical cancer provides precision brachytherapy by simultaneously visualizing the irradiated volumes and spatial relationships of the anatomical structures of the pelvis, the introduced endostats and the corresponding volumetric dose distributions.

Low-cost framework for 3D reconstruction and track detection of the railway network using video data

3D reconstruction from video data is proposed for mapping the Egyptian railway network. A generic framework is proposed and applied based on video data. This framework mainly passes through three main stages. The pre-processing stage contains system implementation, data acquisition, the keyframes extraction, and camera calibration. The second stage is the generation of the 3D reconstruction models from a set of two-dimensional images. This stage is mainly based on the feature correspondence process and Structure from Motion (SfM) techniques. The 3D coordinates of the feature points and relative camera motion parameters are refined using the bundle adjustment algorithms to generate an accurate 3D reconstruction model. Many models are reconstructed based on different parameters/scenarios to evaluate the most accurate model. The best 3D model was selected and used to generate an orthogonal photo. Finally, the third stage is the track detection framework which uses the generated orthogonal image for automatic detection of railway tracks. The framework is tested on a dataset of a railway network with an approximate length of 20 km. The accuracy of the results is compared with field survey data conducted in the same area using conventional surveying instruments. From the results, the recommended train speed for reasonable video data collection is 10 kph with an overlap percentage of 95 %. The RMSE of the extracted track width is 0.03 m, and it contributes to the efficiency of our approach in the reconstruction of 3D surfaces compared to the recommended procedures in traditional techniques.

Commissioning and validation of a novel commercial TPS for ocular proton therapy.

Until today, the majority of ocular proton treatments worldwide were planned with the EYEPLAN treatment planning system (TPS). Recently, the commercial, computed tomography (CT)-based TPS for ocular proton therapy RayOcular was released, which follows the general concepts of model-based treatment planning approach in conjunction with a pencil-beam-type dose algorithm (PBA). To validate RayOcular with respect to two main features: accurate geometrical representation of the eye model and accuracy of its dose calculation algorithm in combination with an Ion Beam Applications (IBA) eye treatment delivery system. Different 3D-printed eye-ball-phantoms were fabricated to test the geometrical representation of the corresponding CT-based model, both in orthogonal 2D images for X-ray image overlay and in fundus view overlaid with a funduscopy. For the latter, the phantom was equipped with a lens matching refraction of the human eye. Funduscopy was acquired in a Zeiss Claus 500 camera. Tantalum clips and fiducials attached to the phantoms were localized in the TPS model, and residual deviations to the actual position in X-ray images for various orientations of the phantom were determined, after the nominal eye orientation was corrected in RayOcular to obtain a best overall fit. In the fundus view, deviations between known and displayed distances were measured. Dose calculation accuracy of the PBA on a 0.2mm grid was investigated by comparing between measured lateral and depth-dose profiles in water for various combinations of range, modulation, and field-size. Ultimately, the modeling of dose distributions behind wedges was tested. A 1D gamma-test was applied, and the lateral and distal penumbra were further compared. Average residuals between model clips and visible clips/fiducials in orthogonal X-ray images were within 0.3mm, including different orientations of the phantom. The differences between measured distances on the registered funduscopy image in the RayOcular fundus view and the known ground-truth were within 1mm up to 10.5mm distance from the posterior pole. No clear benefit projection of either polar mode or camera mode could be identified, the latter mimicking camera properties. Measured dose distributions were reproduced with gamma-test pass-rates of >95% with 2%/0.3mm for depth and lateral profiles in the middle of spread-out Bragg-peaks. Distal falloff and lateral penumbra were within 0.2mm for fields without a wedge. For shallow depths, the agreement was worse, reaching pass-rates down to 80% with 5%/0.3mm when comparing lateral profiles in air. This is caused by low-energy protons from a scatter source in the IBA system not modeled by RayOcular. Dose distributions modified by wedges were reproduced, matching the wedge-induced broadening of the lateral penumbra to within 0.4mm for the investigated cases and showing the excess dose within the field due to wedge scatter. RayOcular was validated for its use with an IBA single scattering delivery nozzle. Geometric modeling of the eye and representation of 2D projections fulfill clinical requirements. The PBA dose calculation reproduces measured distributions and allows explicit handling of wedges, overcoming approximations of simpler dose calculation algorithms used in other systems.

Patient Reported Outcomes after Frameless Virtual Cone Radiosurgery for Trigeminal Neuralgia

<h3>Purpose/Objective(s)</h3> The Virtual Cone is an MLC delivery sequence designed to mimic the dose distribution of a physical cone. In this prospective trial we assessed the efficacy and patient reported outcomes of frameless MLC-based radiosurgery for trigeminal neuralgia. <h3>Materials/Methods</h3> RAD1501 (ClinicalTrials.gov Identifier: NCT02708810) started enrollment in November 2015. Patients had a BNI score of IV or V and MRI to demonstrate no intracranial lesion to explain symptoms. All patients received 80-85 Gy to Dmax to a 5mm target at the dorsal root entry zone. All patients were treated in an Encompass immobilization device. Image guidance started with orthogonal kV images followed by CBCT. During treatment, the patient position was continuously monitored with optical surface imaging. The primary endpoint was an improvement in the Barrow Neurologic Index from IV/V to III or less. Secondary outcomes were assessed by comparing the mean scores of 8 different scales of SF-36 questionnaire at 3 months and 6 months. <h3>Results</h3> At the time of analysis, 44 patients completed treatment with clinical follow up exceeding 6 months. The mean age was 65.8 years (range 35-94 years). Sex distribution was 56.8 % females and 43.2% males. Post treatment BNI at 3 months was 30.8% BNI of 4, 38.5% BNI of 3, 11.2% BNI of 2 and 19.3% with BNI of 1. Post treatment BNI at 6 months was 19% with BNI of 4, 52.4% BNI of 3, 4.8% BNI of 2 and 23.8% with BNI of 1. Regarding secondary outcomes measured by SF-36 questionnaire there was a statistically significant improvement at 3 months in role limitation due to physical health (<i>p=0.023)</i>, energy (<i>p= 0.014)</i>, emotion wellbeing (<i>p=0.002)</i>, social functioning (<i>p=0.008)</i>, and pain (<i>p<0.001).</i> At 6 month there was statistically significant improvement in energy (<i>p= 0.002)</i>, emotion wellbeing (<i>p<0.001)</i>, social functioning (<i>p=0.030)</i>, and pain (<i>p=<0.001).</i> At 3 months 27/27 (100%) of patients were satisfied with the procedure and would recommend it to family member or friend. <h3>Conclusion</h3> From our initial analysis frameless virtual cone stereotactic radiosurgery for trigeminal neuralgia appears to provide similar outcomes in pain relief (∼80% at 6 months) when compared to stereotactic radiosurgery as demonstrated in previous studies. Patient quality of life improves and there is high satisfaction with the procedure.

Initial Experience with Real-Time Gated Proton Therapy (RGPT) in the Definitive Treatment of Prostate Cancer

<h3>Purpose/Objective(s)</h3> RGPT has the potential to provide instantaneous feedback for intrafraction target motion to maximize patient safety and inform optimal treatment planning. Our purpose was to report our early experience with RGPT and to summarize intrafraction target motion and implications on treatment delivery, including added time, for a pilot population of prostate cancer patients before expansion to other sites with greater motion. <h3>Materials/Methods</h3> RGPT with a proton beam therapy system was commissioned and implemented at our proton center, representing the first experience in the United States. Patients being treated definitively with proton therapy to the prostate ± seminal vesicles-only and who had successful placement of platinum fiducial markers and hydrogel spacer were assessed on a prospective IRB-approved registry. Orthogonal fluoroscopic imaging was taken at a pulse rate of one image per second during fractionated proton therapy to assess positioning of a pre-selected fiducial marker. Thresholds to guide beam-on and beam-off gating were determined for a matching score component (based on quality of the image tracking and the guiding template) and for a shift tolerance component (based on our typical set-up uncertainty robustness of 3mm). The shift tolerances were defined as 2mm for left-right and anterior-posterior, and 3mm for superior-inferior. Each patient was treated with single-field optimization planning and a two lateral beam setup. <h3>Results</h3> To date, 8 prostate cancer patients have been treated with a total of 152 RGPT-guided fields. The table details the mean, maximum, standard deviation, and estimated upper 90^th percentile of directional shifts for the entire dataset for 1) the total RGPT-tracked treatment time and 2) the beam-on time as gated by RGPT tracking. On average, the initial imaging to set up the RGPT system required 2.5 seconds per field. The average total treatment time during RGPT guidance, including periods of both "on" and "off" gating, was 41.3 seconds per field. RGPT tracking triggered "beam-off" for an average of 7.8 seconds per field, increasing the total treatment delivery time by 20.2% across the 152 fields. <h3>Conclusion</h3> RGPT successfully tracked intrafraction fiducial marker motion during proton therapy for prostate cancer, while having minimal impact on total treatment time. Further aggregated RGPT-derived data may help to inform optimal treatment planning parameters such as robust optimization.

A small footprint couch-top support device for image-guided radiotherapy of heavy patients.

Patients with body weights close to or above 400 lbs present unique challenges in radiation therapy since the weight limit of most treatment couches decreases as the couch-top extends toward the treatment gantry. The purpose of this work was to develop a small footprint couch-top support platform to safely perform image-guided radiotherapy (IGRT) for extremely heavy patients. One way to protect the couch-top from damage and prevent a catastrophic breakdown is to provide additional support as the couch extends toward the treatment gantry. To allow a maximal range of gantry movement, a small-footprint adjustable jack stand, placed underneath the couch-top, was chosen and modified from a commercial jack stand (with 1100 lbs capacity). The couch could be easily extended longitudinally and laterally with a modified 8-ball-transfer plate mounted at the top. The operation of a couch-top support platform was used for two heavy patients after phantom testing. kV and MV imaging options and ranges were quantified. The custom-constructed couch-top support platform was found to provide stable support with smooth couch shifts. The small footprint allowed gantry rotation from 133° to 227°, which would allow both fixed beam radiotherapy and partial-arc volumetric modulated arc therapy (VMAT). For IGRT, orthogonal 2D kV-kV image pairs with source angles of 40o and 130o were acquired and tested successfully. With the support platform, two clinical cases with patient weights greater than 415 lbs were successfully treated with image-guided partial arc VMAT radiotherapy. The study demonstrated the safety and efficiency of using this new couch-top support platform to prevent couch failure from treating heavy patients. A new couch-top support platform has been designed, assembled, and tested for IGRT. The new support platform is easy to use, cost-effective, and allows extremely heavy patients to be treated safely and robustly with IGRT and VMAT.

Evaluation of intrafraction couch shifts for proton treatment delivery in head-and-neck cancer patients: Toward optimal imaging frequency.

Treatment planning for head-and-neck (H&N) cancer, in particular oropharynx, nasopharynx, and paranasal sinus cases, at our center requires noncoplanar proton beams due to the complexity of the anatomy and target location. Targeting accuracy for all beams is carefully evaluated by using image guidance before delivering proton beam therapy (PBT). In this study, we analyzed couch shifts to evaluate whether imaging is required before delivering each field with different couch angles. After the Institutional Review Board approval, a retrospective analysis was performed on data from 28 H&N patients treated with PBT. Each plan was made with two-to-three noncoplanar and two-to-three coplanar fields. Cone-beam computed tomography and orthogonal kilovoltage (kV) images were acquired for setup and before delivering each field, respectively. The Cartesian (longitudinal, vertical, and lateral) and angular (pitch and roll) shifts for each field were recorded from the treatment summary on the first two fractions and every subsequent fifth fraction. A net magnitude of the three-dimensional (3D) shift in Cartesian coordinates was calculated, and a 3D vector was created from the 6 degrees of freedom coordinates for transforming couch shifts in the system coordinate to the beam's-eye view. A total of 3219 Cartesian and 2146 angular shift values were recorded for 28 patients. Of the Cartesian shifts, 2069 were zero (64.3%), and 1150 (35.7%) were nonzero (range, -7 to 11mm). Of the angular shifts, 1034 (48.2%) were zero, and 1112 (51.8%) were nonzero (range, -3.0° to 3.2°). For 17 patients, the couch shifts increased toward the end of the treatment course. We also found that patients with higher body mass index (BMI) presented increased net couch shifts (p<0.001). With BMI<27, all overall net shift averages were <2mm, and overall maximum net shifts were <6mm. These results confirm the need for orthogonal kV imaging before delivering each field of H&N PBT at our center, where a couch rotation is involved.

How Resilient Are Deep Learning Models in Medical Image Analysis? The Case of the Moment-Based Adversarial Attack (Mb-AdA).

In the past years, deep neural networks (DNNs) have become popular in many disciplines such as computer vision (CV). One of the most important challenges in the CV area is Medical Image Analysis (MIA). However, adversarial attacks (AdAs) have proven to be an important threat to vision systems by significantly reducing the performance of the models. This paper proposes a new black-box adversarial attack, which is based οn orthogonal image moments named Mb-AdA. Additionally, a corresponding defensive method of adversarial training using Mb-AdA adversarial examples is also investigated, with encouraging results. The proposed attack was applied in classification and segmentation tasks with six state-of-the-art Deep Learning (DL) models in X-ray, histopathology and nuclei cell images. The main advantage of Mb-AdA is that it does not destroy the structure of images like other attacks, as instead of adding noise it removes specific image information, which is critical for medical models’ decisions. The proposed attack is more effective than compared ones and achieved degradation up to 65% and 18% in terms of accuracy and IoU for classification and segmentation tasks, respectively, by also presenting relatively high SSIM. At the same time, it was proved that Mb-AdA adversarial examples can enhance the robustness of the model.

Effectiveness of Remote PPG Construction Methods: A Preliminary Analysis.

The contactless recording of a photoplethysmography (PPG) signal with a Red-Green-Blue (RGB) camera is known as remote photoplethysmography (rPPG). Studies have reported on the positive impact of using this technique, particularly in heart rate estimation, which has led to increased research on this topic among scientists. Therefore, converting from RGB signals to constructing an rPPG signal is an important step. Eight rPPG methods (plant-orthogonal-to-skin (POS), local group invariance (LGI), the chrominance-based method (CHROM), orthogonal matrix image transformation (OMIT), GREEN, independent component analysis (ICA), principal component analysis (PCA), and blood volume pulse (PBV) methods) were assessed using dynamic time warping, power spectrum analysis, and Pearson’s correlation coefficient, with different activities (at rest, during exercising in the gym, during talking, and while head rotating) and four regions of interest (ROI): the forehead, the left cheek, the right cheek, and a combination of all three ROIs. The best performing rPPG methods in all categories were the POS, LGI, and OMI methods; each performed well in all activities. Recommendations for future work are provided.

Local optimum underwater polarization imaging enhancement based on connected domain prior

Estimating the degree of polarization (DOP) of target signal light () is a critical step in the underwater polarization differential imaging process. It is challenging to accurately estimate the of objects containing multiple attributes in the imaging range. Inspired by the optical polarization characteristics of objects with different DOP, we find a prior of connected domains (CDs) of objects in polarization orthogonal images. Based on this prior, the CDs of low-DOP (LDOP) and high-DOP (HDOP) objects are extracted from the darkest image () and brightest image () without additional image acquisition. The of LDOP and HDOP objects in the scene can be estimated through the extracted CDs, which realizes the transition from global optimum estimation in the traditional method to local optimum estimation. According to the estimated , the enhanced results of the entire image marked by can be obtained after fusing the enhanced results of the objects in each CD. Experimental results of simple and complex combination scenes of objects show that our method can recognize CDs of LDOP and HDOP objects. We obtained the enhanced results of the entire image, verified the feasibility and robustness of our method, and found that the CDs of LDOP and HDOP objects could be effectively extracted in up to NTU of turbid water. The local optimum improves the application ability of underwater polarization imaging in unpredictable scenarios, while the enhanced result of target labeling has potential application value in underwater target recognition and classification.