Aftereffect of perceived motion trajectories.

If our visual system has a distinct computational process for motion trajectories, such a process may minimize redundancy and emphasize variation in object trajectories by adapting to the current statistics. Our experiments show that after adaptation to multiple objects traveling along trajectories with a common tilt, the trajectory of an object was perceived as tilting on the repulsive side. This trajectory aftereffect occurred irrespective of whether the tilt of the adapting stimulus was physical or an illusion from motion-induced position shifts and did not differ in size across the physical and illusory conditions. Moreover, when the perceived and physical tilts competed during adaptation, the trajectory aftereffect depended on the perceived tilt. The trajectory aftereffect transferred between hemifields and was not explained by motion-insensitive orientation adaptation or attention. These findings provide evidence for a trajectory-specific adaptable process that depends on higher-order representations after the integration of position and motion signals.

Computerized Segmentation Method for Nonmasses on Breast DCE-MRI Images Using ResUNet++ with Slice Sequence Learning and Cross-Phase Convolution.

The purpose of this study was to develop a computerized segmentation method for nonmasses using ResUNet++ with a slice sequence learning and cross-phase convolution to analyze temporal information in breast dynamic contrast material-enhanced magnetic resonance imaging (DCE-MRI) images. The dataset consisted of a series of DCE-MRI examinations from 54 patients, each containing three-phase images, which included one image that was acquired before contrast injection and two images that were acquired after contrast injection. In the proposed method, the region of interest (ROI) slice images are first extracted from each phase image. The slice images at the same position in each ROI are stacked to generate a three-dimensional (3D) tensor. A cross-phase convolution generates feature maps with the 3D tensor to incorporate the temporal information. Subsequently, the feature maps are used as the input layers for ResUNet++. New feature maps are extracted from the input data using the ResUNet++ encoders, following which the nonmass regions are segmented by a decoder. A convolutional long short-term memory layer is introduced into the decoder to analyze a sequence of slice images. When using the proposed method, the average detection accuracy of nonmasses, number of false positives, Jaccard coefficient, Dice similarity coefficient, positive predictive value, and sensitivity were 90.5%, 1.91, 0.563, 0.712, 0.714, and 0.727, respectively, larger than those obtained using 3D U-Net, V-Net, and nnFormer. The proposed method achieves high detection and shape accuracies and will be useful in differential diagnoses of nonmasses.

Potential Usefulness a Coronal View using an Automated Breast Ultrasound System in Detecting Breast Lesions.

Objective: An automated breast ultrasound system (ABUS) combined with screening mammography has increased cancer detection rates; however, supplemental ABUS use has increased recall rates. In this study, we aimed to identify an accurate and efficient method of ABUS interpretation and evaluate the potential usefulness of its coronal view versus the conventional transverse view. Materials and Methods: This retrospective observer study included comprised 114 ABUS cases (40 normal, 35 benign, 39 malignant). Ten physicians from multiple institutions interpreted the anonymized coronal and transverse views independently. The observers scored their confidence in the lesion detection for each case using a continuous scale and recorded reading times for each coronal and transverse view interpretation. Free-response receiver operating characteristic analysis was employed to compare detection accuracies between views; a paired t-test was used to compare the average reading times. Results: Detection accuracy did not differ significantly between the coronal and transverse views (figure of merit=0.740 and 0.745, respectively; p = 0.72). However, the average reading time for the coronal view was significantly shorter than that for the transverse view (149.7 vs. 200.3 seconds per case, p = 0.003). Conclusion: The coronal view obtained with the ABUS was useful for interpretation and associated with significantly shorter reading times compared with the conventional transverse view while maintaining breast lesion detection accuracy.

Theoretical and Technical Issues Concerning the Measurement of Alpha Frequency and the Application of Signal Detection Theory: Comment on Buergers and Noppeney (2022).

Classical and recent evidence has suggested that alpha oscillations play a critical role in temporally discriminating or binding successively presented items. Challenging this view, Buergers and Noppeney [Buergers, S., & Noppeney, U. The role of alpha oscillations in temporal binding within and across the senses. Nature Human Behaviour, 6, 732-742, 2022] found that by combining EEG, psychophysics, and signal detection theory, neither prestimulus nor resting-state alpha frequency influences perceptual sensitivity and bias in the temporal binding task. We propose the following four points that should be considered when interpreting the role of alpha oscillations, and especially their frequency, on perceptual temporal binding: (1) Multiple alpha components can be contaminated in conventional EEG analysis; (2) the effect of alpha frequency on perception will interact with alpha power; (3) prestimulus and resting-state alpha frequency can be different from poststimulus alpha frequency, which is the frequency during temporal binding and should be more directly related to temporal binding; and (4) when applying signal detection theory under the assumption of equal variance, the assumption is often incomplete and can be problematic (e.g., the magnitude relationships between individuals in parametric sensitivity may change when converted into nonparametric sensitivity). Future directions, including solutions to each of the issues, are discussed.

Building a decoder of perceptual decisions from microsaccades and pupil size.

Many studies have reported neural correlates of visual awareness across several brain regions, including the sensory, parietal, and frontal areas. In most of these studies, participants were instructed to explicitly report their perceptual experience through a button press or verbal report. It is conceivable, however, that explicit reporting itself may trigger specific neural responses that can confound the direct examination of the neural correlates of visual awareness. This suggests the need to assess visual awareness without explicit reporting. One way to achieve this is to develop a technique to predict the visual awareness of participants based on their peripheral responses. Here, we used eye movements and pupil sizes to decode trial-by-trial changes in the awareness of a stimulus whose visibility was deteriorated due to adaptation-induced blindness (AIB). In the experiment, participants judged whether they perceived a target stimulus and rated the confidence they had in their perceptual judgment, while their eye movements and pupil sizes were recorded. We found that not only perceptual decision but also perceptual confidence can be separately decoded from the eye movement and pupil size. We discuss the potential of this technique with regard to assessing visual awareness in future neuroimaging experiments.

Pseudocontinuous Arterial Spin Labeling: Clinical Applications and Usefulness in Head and Neck Entities.

As functional magnetic resonance imaging, arterial spin labeling (ASL) techniques have been developed to provide quantitative tissue blood flow measurements, which can improve the performance of lesion diagnosis. ASL does not require contrast agents, thus, it can be applied to a variety of patients regardless of renal impairments and contrast agent allergic reactions. The clinical implementation of head and neck lesions is limited, although, in recent years, ASL has been increasingly utilized in brain lesions. Here, we review the development of the ASL techniques, including pseudocontinuous ASL (pCASL). We compare readout methods between three-dimensional (3D) turbo spin-echo and 2D echo planar pCASL for the clinical applications of pCASL to head and neck lesions. We demonstrate the clinical usefulness of 3D pCASL for diagnosing various entities, including inflammatory lesions, hypervascular lesions, and neoplasms; for evaluating squamous cell carcinoma (SCC) treatment responses, and for predicting SCC prognosis.

Tooth detection for each tooth type by application of faster R-CNNs to divided analysis areas of dental panoramic X-ray images.

This study aimed to propose a computerized method for detecting the tooth region for each tooth type as the initial stage in the development of a computer-aided diagnosis (CAD) scheme for dental panoramic X-ray images. Our database consists of 160 panoramic dental X-ray images obtained from 160 adult patients. To reduce false positives (FPs), the proposed method first extracts a rectangular area including all teeth from a dental panoramic X-ray image with a faster region using a convolutional neural network (Faster R-CNN). From the rectangular area including all teeth, six divided areas are then extracted with Faster R-CNN: top left, top center, top right, bottom left, bottom center, and bottom right. Faster R-CNNs for detecting tooth regions for each tooth type were trained individually for each of the divided areas that narrowed down the target tooth types. By applying these Faster R-CNNs to each divided area, the bounding boxes of each tooth were detected and classified into 32 tooth types. A k-fold cross-validation method with k = 4 was used for training and testing the proposed method. The detection rate for each tooth, number of FPs per image, mean intersection over union for each tooth, and classification accuracy for the 32 tooth types were 98.9%, 0.415, 0.748, and 91.7%, respectively, showing an improvement compared to the application of the Faster R-CNN once to the entire image (98.0%, 1.194, 0.736, and 88.8%).

Semiautomated Segmentation and Volume Measurements of Cervical Carotid High-Signal Plaques Using 3D Turbo Spin-Echo T1-Weighted Black-Blood Vessel Wall Imaging: A Preliminary Study.

Unstable carotid plaques are visualized as high-signal plaques (HSPs) on 3D turbo spin-echo T1-weighted black-blood vessel wall imaging (3D TSE T1-BB VWI). The purpose of this study was to compare manual segmentation and semiautomated segmentation for the quantification of carotid HSPs using 3D TSE T1-BB VWI. Twenty cervical carotid plaque lesions in 19 patients with a plaque contrast ratio of > 1.3 compared to adjacent muscle were studied. Using the mean voxel value for the adjacent muscle multiplied by 1.3 as a threshold value, the semiautomated software exclusively segmented and measured the HSP volume. Manual and semiautomated HSP volumes were well correlated (r = 0.965). Regarding reproducibility, the inter-rater ICC was 0.959 (bias: 24.63, 95% limit of agreement: −96.07, 146.35) for the manual method and 0.998 (bias: 15.2, 95% limit of agreement: −17.83, 48.23) for the semiautomated method, indicating improved reproducibility by the semiautomated method compared to the manual method. The time required for semiautomated segmentation was significantly shorter than that of manual segmentation times (81.7 ± 7.8 s versus 189.5 ± 49.6 s; p < 0.01). The results obtained in this study demonstrate that the semiautomated segmentation method allows for reliable assessment of the HSP volume in patients with carotid plaque lesions, with reduced time and effort for the analysis.

Tumor blood flow and apparent diffusion coefficient histogram analysis for differentiating malignant salivary tumors from pleomorphic adenomas and Warthin's tumors.

We aimed to assess the combined diagnostic value of apparent diffusion coefficient (ADC) and tumor blood flow (TBF) obtained by pseudocontinuous arterial spin labeling (pCASL) for differentiating malignant tumors (MTs) in salivary glands from pleomorphic adenomas (PAs) and Warthin's tumors (WTs). We used pCASL imaging and ADC map to evaluate 65 patients, including 16 with MT, 30 with PA, and 19 with WT. We evaluated all tumors by histogram analyses and compared various characteristics by one-way analysis of variance followed by Tukey post-hoc tests. Diagnostic performance was evaluated by receiver operating characteristic (ROC) curve analysis. There were significant differences in the mean, 50th, 75th, and 90th percentiles of TBF among the tumor types, in the mean TBFs (mL/100 g/min) between MTs (57.47 ± 35.14) and PAs (29.88 ± 22.53, p = 0.039) and between MTs and WTs (119.31 ± 50.11, p < 0.001), as well as in the mean ADCs (× 10-3 mm2/s) between MTs (1.08 ± 0.28) and PAs (1.60 ± 0.34, p < 0.001), but not in the mean ADCs between MTs and WTs (0.87 ± 0.23, p = 0.117). In the ROC curve analysis, the highest areas under the curves (AUCs) were achieved by the 10th and 25th percentiles of ADC (AUC = 0.885) for differentiating MTs from PAs and the 50th percentile of TBF (AUC = 0.855) for differentiating MTs from WTs. The AUCs of TBF, ADC, and combination of TBF and ADC were 0.850, 0.885, and 0.950 for MTs and PAs differentiation and 0.855, 0.814, and 0.905 for MTs and WTs differentiation, respectively. The combination of TBF and ADC evaluated by histogram analysis may help differentiate salivary gland MTs from PAs and WTs.

A novel strategy to develop deep learning for image super-resolution using original ultra-high-resolution computed tomography images of lung as training dataset.

PURPOSE: To improve the image quality of inflated fixed cadaveric human lungs by utilizing ultra-high-resolution computed tomography (U-HRCT) as a training dataset for super-resolution processing using deep learning (SR-DL). MATERIALS AND METHODS: Image data of nine cadaveric human lungs were acquired using U-HRCT. Three different matrix images of U-HRCT images were obtained with two acquisition modes: normal mode (512-matrix image) and super-high-resolution mode (1024- and 2048-matrix image). SR-DL used 512- and 1024-matrix images as training data for deep learning. The virtual 2048-matrix images were acquired by applying SR-DL to the 1024-matrix images. Three independent observers scored normal anatomical structures and abnormal computed tomography (CT) findings of both types of 2048-matrix images on a 3-point scale compared to 1024-matrix images. The image noise values were quantitatively calculated. Moreover, the edge rise distance (ERD) and edge rise slope (ERS) were also calculated using the CT attenuation profile to evaluate margin sharpness. RESULTS: The virtual 2048-matrix images significantly improved visualization of normal anatomical structures and abnormal CT findings, except for consolidation and nodules, compared with the conventional 2048-matrix images (p < 0.01). Quantitative noise values were significantly lower in the virtual 2048-matrix images than in the conventional 2048-matrix images (p < 0.001). ERD was significantly shorter in the virtual 2048-matrix images than in the conventional 2048-matrix images (p < 0.01). ERS was significantly higher in the virtual 2048-matrix images than in the conventional 2048-matrix images (p < 0.01). CONCLUSION: SR-DL using original U-HRCT images as a training dataset might be a promising tool for image enhancement in terms of margin sharpness and image noise reduction. By applying trained SR-DL to U-HRCT SHR mode images, virtual ultra-high-resolution images were obtained which surpassed the image quality of unmodified SHR mode images.

Evaluation of placental oxygenation index using blood oxygen level-dependent magnetic resonance imaging (BOLD-MRI) during normal late pregnancy.

AIM: Noninvasive blood oxygen level-dependent magnetic resonance imaging (BOLD-MRI) has recently been used to evaluate placental oxygenation. However, this method still has unresolved problems, such as long testing times and lack of normal values set. In the present study, we used a shorter protocol in BOLD-MRI and established normal values for placental oxygenation in late pregnancy. METHODS: We recruited 18 healthy singleton pregnant women (>32 weeks of gestation) who had a normal body size before pregnancy and a normal course of pregnancy. They underwent BOLD-MRI with three consecutive 4-min periods of different oxygenation: normoxia (21% O2), hyperoxia (10 L O2/min), and then normoxia. Placental time-activity curves were presented as signal intensity change relative to baseline (ΔR2*). The time from starting maternal oxygen administration to peak ΔR2*. To assess the relationship between peak ΔR2* values and placenta-related parameters and fetal development, the correlation between peak ΔR2*, placental weight, and neonatal birth weight was evaluated using Spearman's rank correlation test. RESULTS: In all cases, the BOLD signal was elevated by maternal oxygen administration, with the peak resolving within 4 min after the end of oxygen administration. Peak ΔR2* and time to peak ΔR2* during oxygenation were 7.99 ± 2.58, and 458.1 ± 73.9 s, respectively. There was a significant correlation between peak ΔR2* and neonatal birth weight (percentile) (r = 0.537, p = .022), and between placental weight and neonatal birth weight (r = 0.769, p < .01). CONCLUSIONS: In all cases, the BOLD signal increased with maternal hyperoxia using this protocol. So, 4 min observation following maternal oxygen administration is sufficient for peak ΔR2* evaluation. These reference values set in this study may be one of the indicators of BOLD signal changes in normal pregnancies after 32 weeks of gestation.

A dynamic noise background reveals perceptual motion extrapolation: The twinkle-goes illusion.

We find that on a dynamic noise background, the perceived disappearance location of a moving object is shifted in the direction of motion. This "twinkle-goes" illusion does not require luminance- or chromaticity-based confusability of the object with the background, or on the amount of background motion energy in the same direction as the object motion. This suggests that the illusion is enabled by the dynamic noise masking the offset transients that otherwise accompany an object's disappearance. While these results are consistent with an anticipatory process that pre-activates positions ahead of the object's current position, additional findings suggest an alternative account: a continuation of attentional tracking after the object disappears. First, the shift increased with speed until over 1.2 revolutions per second (rps), nearing the attentional tracking limit. Second, the shift was greatly reduced when attention was divided between two moving objects. Finally, the illusion was associated with a delay in simple reaction time to the disappearance of the object. We propose that in the absence of offset transients, attentional tracking keeps moving for several tens of milliseconds after the target disappearance, and this causes one to hallucinate a moving object at the position of attention.

Segmentation of teeth in panoramic dental X-ray images using U-Net with a loss function weighted on the tooth edge.

Panoramic dental X-ray imaging is an established method for the diagnosis of dental problems. However, the resolution of panoramic dental X-ray images is relatively low. Thus, early lesions are often overlooked. As the first step in the development of a computer-aided diagnosis scheme for panoramic dental X-ray images, we propose a computerized method for the segmentation of teeth using U-Net with a loss function weighted on the tooth edge. Our database consisted of 162 panoramic dental X-ray images. The training dataset consisted of 102 images, while the remaining 60 images were used as the test dataset. The loss function obtained by the cross entropy (CE) in the entire image is usually used in training U-Net. To improve the segmentation accuracy of the tooth edge, a loss function weighted on the tooth edge is proposed by adding the CE in the tooth edge region to the CE for the entire image. The mean Jaccard index and Dice index for U-Net with the loss function combining the CEs for the entire image and tooth edge were 0.864 and 0.927, respectively, which were significantly larger than those for U-Net with the CE for the entire image (0.802 and 0.890, p < 0.001) and U-Net with the CE for the tooth edge (0.826 and 0.905, p < 0.001). U-Net with the new loss function exhibited a higher segmentation accuracy of the tooth in panoramic dental X-ray images than that obtained by U-Net with the conventional loss function.

Diagnostic performance of coronal view in comparison with transverse view of three-dimensional automated breast ultrasound.

BACKGROUND: Automated breast ultrasound (ABUS) is one of the first ultrasound devices which enables the de-coupling of image acquisition and interpretation. Another feature of ABUS is the coronal view, utilizing three-dimensional volume data reconstructed from two-dimensional transverse images acquired automatically. PURPOSE: To assess the diagnostic performance of coronal view interpretation by comparing it with that of the transverse view. MATERIAL AND METHODS: This was a retrospective, multi-case, observer study using a cancer-enriched dataset of ABUS images at a single institution with approval by an Institutional Review Board. The 100 scan datasets selected between October 2014 and January 2017 included 70 non-cancer cases and 30 malignancies. In the present observer study, two experienced physicians provided their confidence levels regarding the malignancy of each of the 100 scan datasets independently. The reading times for interpretation of coronal and transverse views were recorded. RESULTS: Area under the receiver operating characteristic curves for two observers with the transverse view (0.856) was improved by use of the coronal view (0.917, P = 0.036). The average reading times were 140.4 s with the coronal view and 148.5 s with the transverse view per scan dataset (P = 0.246). CONCLUSION: It is conceivable that the accurate use of the coronal view will lead to improvement in diagnostic performance in breast cancer screening, although this needs to be confirmed with a larger prospective study.

Computer-Aided Diagnosis Scheme for Distinguishing Between Benign and Malignant Masses on Breast DCE-MRI Images Using Deep Convolutional Neural Network with Bayesian Optimization.

Although magnetic resonance imaging (MRI) has a higher sensitivity of early breast cancer than mammography, the specificity is lower. The purpose of this study was to develop a computer-aided diagnosis (CAD) scheme for distinguishing between benign and malignant breast masses on dynamic contrast material-enhanced MRI (DCE-MRI) by using a deep convolutional neural network (DCNN) with Bayesian optimization. Our database consisted of 56 DCE-MRI examinations for 56 patients, each of which contained five sequential phase images. It included 26 benign and 30 malignant masses. In this study, we first determined a baseline DCNN model from well-known DCNN models in terms of classification performance. The optimum architecture of the DCNN model was determined by changing the hyperparameters of the baseline DCNN model such as the number of layers, the filter size, and the number of filters using Bayesian optimization. As the input of the proposed DCNN model, rectangular regions of interest which include an entire mass were selected from each of DCE-MRI images by an experienced radiologist. Three-fold cross validation method was used for training and testing of the proposed DCNN model. The classification accuracy, the sensitivity, the specificity, the positive predictive value, and the negative predictive value were 92.9% (52/56), 93.3% (28/30), 92.3% (24/26), 93.3% (28/30), and 92.3% (24/26), respectively. These results were substantially greater than those with the conventional method based on handcrafted features and a classifier. The proposed DCNN model achieved high classification performance and would be useful in differential diagnoses of masses in breast DCE-MRI images as a diagnostic aid.

Attention updates the perceived position of moving objects.

The information used by conscious perception may differ from that which drives certain actions. A dramatic illusion caused by an object's internal texture motion has been put forward as one example. The motion causes an illusory position shift that accumulates over seconds into a large effect, but targeting of the grating for a saccade (a rapid eye movement) is not affected by this illusion. While this has been described as a dissociation between perception and action, an alternative explanation is that rather than saccade targeting having privileged access to the correct position, a shift of attention that precedes saccades resets the accumulated illusory position shift to zero. In support of this possibility, we found that the accumulation of illusory position shift can be reset by transients near the moving object, creating an impression of the object returning to near its actual position. Repetitive luminance changes of the object also resulted in reset of the accumulation, but less so when attention to the object was reduced by a concurrent digit identification task. Finally, judgments of the object's positions around the time of saccade onset reflected the veridical rather than the illusory position. These results suggest that attentional shifts, including those preceding saccades, can update the perceived position of moving objects and mediate the previously reported dissociation between conscious perception and saccades.

Improving Image Resolution of Whole-Heart Coronary MRA Using Convolutional Neural Network.

Whole-heart coronary magnetic resonance angiography (WHCMRA) permits the noninvasive assessment of coronary artery disease without radiation exposure. However, the image resolution of WHCMRA is limited. Recently, convolutional neural networks (CNNs) have obtained increased interest as a method for improving the resolution of medical images. The purpose of this study is to improve the resolution of WHCMRA images using a CNN. Free-breathing WHCMRA images with 512 × 512 pixels (pixel size = 0.65 mm) were acquired in 80 patients with known or suspected coronary artery disease using a 1.5 T magnetic resonance (MR) system with 32 channel coils. A CNN model was optimized by evaluating CNNs with different structures. The proposed CNN model was trained based on the relationship of signal patterns between low-resolution patches (small regions) and the corresponding high-resolution patches using a training dataset collected from 40 patients. Images with 512 × 512 pixels were restored from 256 × 256 down-sampled WHCMRA images (pixel size = 1.3 mm) with three different approaches: the proposed CNN, bicubic interpolation (BCI), and the previously reported super-resolution CNN (SRCNN). High-resolution WHCMRA images obtained using the proposed CNN model were significantly better than those of BCI and SRCNN in terms of root mean squared error, peak signal to noise ratio, and structure similarity index measure with respect to the original WHCMRA images. The proposed CNN approach can provide high-resolution WHCMRA images with better accuracy than BCI and SRCNN. The high-resolution WHCMRA obtained using the proposed CNN model will be useful for identifying coronary artery disease.

Usefulness of dictionary learning-based processing for improving image quality of sub-millisievert low-dose chest CT: initial experience.

PURPOSE: To develop a dictionary learning (DL)-based processing technique for improving the image quality of sub-millisievert chest computed tomography (CT). MATERIALS AND METHODS: Standard-dose and sub-millisievert chest CT were acquired in 12 patients. Dictionaries including standard- and low-dose image patches were generated from the CT datasets. For each patient, DL-based processing was performed for low-dose CT using the dictionaries generated from the remaining 11 patients. This procedure was repeated for all 12 patients. Image quality of normal thoracic structures on the processed sub-millisievert CT images was assessed with a 5-point scale (5 = excellent, 1 = very poor). Lung lesion conspicuity was also assessed on a 5-point scale. RESULTS: Image noise on sub-millisievert CT was significantly decreased with DL-based image processing (48.5 ± 13.7 HU vs 20.4 ± 7.9 HU, p = 0.0005). Image quality of lung structures was significantly improved with DL-based method (middle level of lung, 2.25 ± 0.75 vs 2.92 ± 0.79, p = 0.0078). Lung lesion conspicuity was also significantly improved with DL-based technique (solid nodules, 3.4 ± 0.6 vs 2.7 ± 0.6, p = 0.0273). CONCLUSION: Image quality and lesion conspicuity on sub-millisievert chest CT images may be improved by DL-based post-processing.