Slice Level Research Articles

8740 Visceral Adipose Tissue Assessment in Patients With Mild Autonomous Cortisol Secretion Using Dual-Energy X-Ray Absorptiometry: A Cross-Sectional Study

Abstract Disclosure: J. Saini: None. C.D. Zhang: None. V. Fell: None. E. Atkinson: None. S. Achenbach: None. I. Bancos: None. Background: Several studies employing computed tomography (CT) scans for body composition measurements have reported lower muscle mass and a higher proportion of visceral fat in patients with mild autonomous cortisol secretion (MACS) compared to referent subjects. While a CT scan assesses visceral adipose tissue (VAT) at a single slice level (L3), no studies have evaluated the overall VAT in patients with MACS using Dual-Energy X-ray absorptiometry (DXA). Objective: To determine the impact of MACS on VAT using DXA scan as compared to referent subjects. Methods: We conducted a single-center cross-sectional study of adults with 75 MACS and 75 referent subjects. MACS was defined as serum cortisol>1.8 mcg/dL after 1 mg dexamethasone suppression test in patients with adrenal adenomas and no features of overt hypercortisolism. Referent subjects were included if they had abdominal imaging and had no adrenal disorders. All the participants underwent a full body DXA scan between 04/2019 and 07/2022. We applied the CoreScan software to estimate the (1) total lean and fat mass (2) visceral fat mass and volume (3) lean android, fat android, and android mass, and compared measurements of patients and referent subjects using linear regression models adjusted for sex and BMI. Results: A total of 75 patients with MACS (median age 57.0 years, IQR 47.8-65.6, 67% women) and 75 referent subjects (median age 57.6 years, IQR 49.1-65.8, 67% women) were included. Patients had a higher BMI compared to referent subjects (median 31.7 kg/m2, IQR 27.4-37.9 vs 28.7, IQR, 25.4-33.8, P=0.004). The median post-DST cortisol in patients was 3.1 mcg/dL (IQR, 2.2-5.2). When compared to referent subjects, and after adjustment for sex and BMI, patients had a similar total body fat (estimate=846, p-value=0.28) and lean mass (estimate=-201, p-value=0.83), similar android fat mass (estimate=59, p-value=0.64) but lower gynoid fat mass (estimate=400, 0.007). Notably, sex and BMI-adjusted analysis also showed that patients had a higher VAT fat mass (estimate= -225, p-value=0.084) and a higher VAT fat mass/total mass ratio (estimate= -0.007, p-value=0.026). Conclusions: When adjusted for BMI and sex, patients with MACS do not demonstrate differences in overall fat and lean mass. However, we showed that MACS is associated with increased visceral adiposity, a known cardiovascular risk factor. Body composition assessment by DXA could be a safer and less expensive alternative in the evaluation and follow-up of patients with MACS. Presentation: 6/2/2024

Potential of the Thoracoepigastric Vein as a Drainage Vein in Breast Reconstruction With a Free Flap: An Anatomical Study Using Computed Tomography.

The recipient vessel choice is very important when performing free-flap breast reconstructions. Usually, the concomitant vein of the recipient artery is anastomosed, and mismatches in the diameter are occasionally observed. We consider the thoracoepigastric vein (TEV) as a potential useful recipient vein. The use of the TEV is not a novel technique. It has been used by surgeons for free-flap anastomoses in the axillary region, but usually as an anastomotic site for the second vein. However, anatomical findings such as TEV diameter, its deficiency rate, and influence on mastectomy are not clear. In this study, computed tomography (CT) was performed to evaluate the use of the TEV as a recipient vein for breast reconstruction. The medical records of patients who underwent breast reconstruction with free-flap transfer were retrospectively evaluated. In most cases, CT was performed using a tissue expander inserted after mastectomy. The TEV was considered suitable if its diameter on the mastectomy side was >1 mm and the vertical distance from the most distal slice level where the TEV could be seen at the third costal cartilage height of the parasternal region was <50 mm. Sixty-six sides of 33 patients were evaluated. The mean age of the patients was 49.3 ± 5.1 years. The TEV was used in five flaps. TEV diameters were not significantly different between the left and right sides (1.85 ± 0.53 vs. 1.82 ± 0.6, respectively, p = 0.836). On the mastectomy side, eight (25%) TEVs were injured, but only three (8.6%) were considered unavailable. On the healthy side, 96.4% TEVs were available. TEVs may be a good choice as vein anastomosis targets for breast reconstruction with a free flap because of their relatively large vessel diameter. UMIN-CTR: R000061573.

Artificial intelligence-based pulmonary embolism classification: Development and validation using real-world data.

This paper presents an artificial intelligence-based classification model for the detection of pulmonary embolism in computed tomography angiography. The proposed model, developed from public data and validated on a large dataset from a tertiary hospital, uses a two-dimensional approach that integrates temporal series to classify each slice of the examination and make predictions at both slice and examination levels. The training process consists of two stages: first using a convolutional neural network InceptionResNet V2 and then a recurrent neural network long short-term memory model. This approach achieved an accuracy of 93% at the slice level and 77% at the examination level. External validation using a hospital dataset resulted in a precision of 86% for positive pulmonary embolism cases and 69% for negative pulmonary embolism cases. Notably, the model excels in excluding pulmonary embolism, achieving a precision of 73% and a recall of 82%, emphasizing its clinical value in reducing unnecessary interventions. In addition, the diverse demographic distribution in the validation dataset strengthens the model's generalizability. Overall, this model offers promising potential for accurate detection and exclusion of pulmonary embolism, potentially streamlining diagnosis and improving patient outcomes.

Analysis of culture and RNA isolation methods for precision-cut liver slices from cirrhotic rats

Precision-cut liver slices (PCLS) are increasingly used as a model to investigate anti-fibrotic therapies. However, many studies use PCLS from healthy animals treated with pro-fibrotic stimuli in culture, which reflects only the early stages of fibrosis. The effects of different culture conditions on PCLS from cirrhotic animals has not been well characterized and there is no consensus on optimal methods. In this study, we report a method for the collection and culture of cirrhotic PCLS and compare the effect of common culture conditions on viability, function, and gene expression. Additionally, we compared three methods of RNA isolation and identified a protocol with high yield and purity. We observed significantly increased albumin production when cultured with insulin-transferrin-selenium and dexamethasone, and when incubated on a rocking platform. Culturing with insulin-transferrin-selenium and dexamethasone maintained gene expression closer to the levels in fresh slices. However, despite stable viability and function up to 4 days, we found significant changes in expression of key genes by day 2. Interestingly, we also observed that cirrhotic PCLS maintain viability in culture longer than slices from healthy animals. Due to the influence of matrix stiffness on fibrosis and hepatocellular function, it is important to evaluate prospective anti-fibrotic therapies in a platform that preserves tissue biomechanics. PCLS from cirrhotic animals represent a promising tool for the development of treatments for chronic liver disease.

Protecting Prostate Cancer Classification From Rectal Artifacts via Targeted Adversarial Training.

Magnetic resonance imaging (MRI)-based deep neural networks (DNN) have been widely developed to perform prostate cancer (PCa) classification. However, in real-world clinical situations, prostate MRIs can be easily impacted by rectal artifacts, which have been found to lead to incorrect PCa classification. Existing DNN-based methods typically do not consider the interference of rectal artifacts on PCa classification, and do not design specific strategy to address this problem. In this study, we proposed a novel Targeted adversarial training with Proprietary Adversarial Samples (TPAS) strategy to defend the PCa classification model against the influence of rectal artifacts. Specifically, based on clinical prior knowledge, we generated proprietary adversarial samples with rectal artifact-pattern adversarial noise, which can severely mislead PCa classification models optimized by the ordinary training strategy. We then jointly exploited the generated proprietary adversarial samples and original samples to train the models. To demonstrate the effectiveness of our strategy, we conducted analytical experiments on multiple PCa classification models. Compared with ordinary training strategy, TPAS can effectively improve the single- and multi-parametric PCa classification at patient, slice and lesion level, and bring substantial gains to recent advanced models. In conclusion, TPAS strategy can be identified as a valuable way to mitigate the influence of rectal artifacts on deep learning models for PCa classification.

Presence and distribution of voids after using the single cone obturation technique with different sealer placement methods in canals with an isthmus

Background/purposeThe obturation of canals with irregular structures is still a challenge for single cone obturation technique (SC). The purpose of this study was to evaluate the presence and distribution of voids using SC with different sealer placement methods in the canal with a simulated band-shaped isthmus. Materials and methods3D-printed root canal models with band-shaped isthmuses were randomly divided into four groups according to different obturation methods. Group 1: sealer placement by single cone passively (SCP); Group 2: bi-directional spiral-supported sealer placement (BS); Group 3: ultrasound-supported sealer placement (U). Group 4: vertical compaction obturation (VC). In each group, 10 of 14 models were sliced and the remaining four were scanned by micro-CT. The percentage area of voids (PAV) and the percentage volume of voids (PVV) of fillings were calculated. ResultsAt all slice levels, using BS and U to support sealer placement reduced voids with an average PAV of 21%, and in the VC and SCP groups were 33% and 45% respectively. Based on the micro-CT scans, nearly half of the porosity decreased by the BS and ultrasound in the isthmus with PVV of 25% and 29% respectively, compared with 46% in the SCP group. However, in the main canal, when the PVV was 22% in the SCP group, the porosity decreased to 14% in the U group and 18% in the BS group. ConclusionBi-directional spiral or ultrasound-supported sealer placement can improve the performance of single cone obturation in canals with an isthmus.

End-to-End Semi-Supervised Opportunistic Osteoporosis Screening Using Computed Tomography.

Osteoporosis is the most common metabolic bone disease and can cause fragility fractures. Despite this, screening utilization rates for osteoporosis remain low among populations at risk. Automated bone mineral density (BMD) estimation using computed tomography (CT) can help bridge this gap and serve as an alternative screening method to dual-energy X-ray absorptiometry (DXA). The feasibility of an opportunistic and population agnostic screening method for osteoporosis using abdominal CT scans without bone densitometry phantom-based calibration was investigated in this retrospective study. A total of 268 abdominal CT-DXA pairs and 99 abdominal CT studies without DXA scores were obtained from an oncology specialty clinic in the Republic of Korea. The center axial CT slices from the L1, L2, L3, and L4 lumbar vertebrae were annotated with the CT slice level and spine segmentation labels for each subject. Deep learning models were trained to localize the center axial slice from the CT scan of the torso, segment the vertebral bone, and estimate BMD for the top four lumbar vertebrae. Automated vertebra-level DXA measurements showed a mean absolute error (MAE) of 0.079, Pearson's r of 0.852 (P<0.001), and R2 of 0.714. Subject-level predictions on the held-out test set had a MAE of 0.066, Pearson's r of 0.907 (P<0.001), and R2 of 0.781. CT scans collected during routine examinations without bone densitometry calibration can be used to generate DXA BMD predictions.

Deep multiscale convolutional feature learning for intracranial hemorrhage classification and weakly supervised localization

ObjectiveThis study evaluated the performance of attentional fusion model-based multiscale features in classifying intracerebral hemorrhage and the localization of bleeding focus based on weakly supervised target localization. MethodsA publicly available dataset provided by the American College of Neuroradiology (ASNR) was used, consisting of 750,000 computed tomography (CT) scans of the brain, manually marked by radiologists for intracranial hemorrhage and five hemorrhage subtypes. A multiscale feature classification and weakly supervised localization framework based on an attentional fusion mechanism were applied, which could be annotated at the slice level and provided intracranial hemorrhage classification and hemorrhage focus localization. ResultsThe designed framework achieved excellent performance for classification and localization. The area under the curve (AUC) for predicting bleeding was 0.973. High AUC values were observed for the five hemorrhage subtypes (epidural AUC = 0.891, subdural AUC = 0.991, subarachnoid AUC = 0.983, intraventricular AUC = 0.995, intraparenchymal AUC = 0.990). This model outperformed the average entry-level radiology trainee compared to previously reported data. ConclusionThe designed method quickly and accurately detected intracerebral hemorrhage, classifying hemorrhage subtypes and locating bleeding points with image-level annotation alone. The results indicate that this framework can significantly reduce diagnostic time while improving the detection of intracerebral hemorrhage in emergencies. It can thus be integrated into the diagnostic radiology workflow in the future.

Atractylenolide II ameliorates myocardial fibrosis and oxidative stress in spontaneous hypertension rats.

Hypertension is a well-recognized risk factor for cardiovascular, which is also a critical factor in causing myocardial fibrosis (MF). The study aimed to explore the effect of Atractylenolide II (ATL-II) on MF and oxidative stress in spontaneous hypertension rats (SHR). The body weight of rats after injection of ATL-II was quantitatively analyzed. The left ventricular function of SHR was evaluated by Echocardiographic. HE staining, Masson trichrome staining, left ventricular mass index (LVMI) and immunofluorescence was applied to investigate the effects of ATL-II on MF. RT qPCR was used to detect the Collagen I, α-SMA, Fibronectin, and Vimentin mRNA expression levels in myocardial slices. The effect ATL-II on cardiomyocyte apoptosis was detected by TUNEL staining and western blot. An immunohistochemistry assay was conducted to detect α-SMA protein and TGF-β1 protein. The contents of H2O2, GSH-PX, SOD, and MDA were measured by colorimetry. ATL-II could dose-dependently improve the BW of SHRs (P< 0.05) and enhance myocardial function. Moreover, ATL-II effectively reduced cardiomyocyte apoptosis in SHRs. Alternatively, ATL-II could inhibit the Collagen I, α-SMA, Fibronectin, and Vimentin mRNA and protein expression levels in SHRs. ATL-II could ameliorate oxidative stress by improving the activities of SOD and GSH-PX and lowering the contents of H2O2 and MDA in ATL-II-treated SHRs, which reach about 80%. ATL-II could exert an inhibiting effect on MF and oxidative stress in SHRs. Hence, ATL-II may hold promise for the treatment of MF and oxidative stress in Spontaneous Hypertension.

Febrile seizures cause a decrease in calcium-permeable AMPA receptors at synapses of rat cortical and hippocampal pyramidal neurons

Febrile seizures (FS) are a prevalent childhood neurological disorder that can result in lasting functional alterations in neural networks, contributing to the onset of epilepsy and cognitive impairment [1]. Higher levels of calcium-permeable AMPA receptors (CP-AMPARs), which lack the GluA2 subunit, are detected in the hippocampus and cortex at an early age. CP-AMPARs are involved in various plastic changes within the central nervous system (CNS), including regular physiological processes (synaptic plasticity) and various pathological conditions. Such changes occur when these receptors are included in the neurons’ membrane, where they are not typically expressed. CP-AMPARs were demonstrated to incorporate into synapses during seizures [2]. The effect of FSs on CP-AMPAR expression is uncertain, and the resultant alterations in neuronal network function are unclear. The aim of this study was to determine whether the proportion of CP-AMPARs at synapses of pyramidal neurons in the rat entorhinal cortex and hippocampus alters immediately (at 15 min) and 48 h post FS. Ten-day-old rats were exposed to a stream of warm air (46 °C) for 30 minutes to induce hyperthermia, resulting in the development of FS. Only animals with FS that lasted for a minimum of 15 minutes were included in the study. The control group was comprised of littermates removed from the dam for an equivalent period but kept at room temperature. Isolated pyramidal neurons were used to determine the proportion of CP-AMPARs in the hippocampus. AMPAR-mediated currents were induced by application of 100 μM kainate. Excitatory postsynaptic currents (EPSCs) were evoked by extracellular stimulation in the entorhinal cortex. The antagonist IEM-1460 was used to selectively block CP-AMPARs. The rectification index of AMPA-mediated EPSCs was calculated to better assess the contribution of CP-AMPARs. Neurons expressing CP-AMPARs were visualized using the kainate-induced cobalt uptake method. Brain slices were stimulated with kainate while AR-5 and TTX were present. The DNQX blocker was used for the determination that the influx of Co2+ was mediated by AMPARs. Basal synaptic transmission was assessed by recording field postsynaptic responses in the hippocampus stimulated by Shaffer collaterals at different current strengths. FS induced a rapid decrease in the levels of CP-AMPARs on the membranes of pyramidal neurons in the hippocampus. As a result, 15 min after FS, IEM-1460 caused significantly less blockade of kainate-evoked current in neurons isolated from rat hippocampus compared to control (22% vs. 14%, p 0.05). A similar finding was observed for EPSCs evoked extracellularly in the entorhinal cortex, with a frequency of 10% in the FS group compared to 3% in the control group (p 0.05). Furthermore, the FS group’s neurons exhibited a significantly greater rectification index of EPSCs compared to the control group’s neurons. However, two days post-FS, no significant differences existed between the two groups. This observation may be attributed to the rapid decrease in the proportion of CP-AMPARs in pyramidal neurons at this stage. The cobalt uptake method supported electrophysiological findings, revealing higher staining levels in the CA1 field of the hippocampus and entorhinal cortex of control rats compared to FS rats. The observed effect surfaced 15 minutes after FS, and no divergences emerged between the groups after two days. Although the proportion of CP-AMPARs was reduced, basal neurotransmission levels in brain slices obtained from rats that underwent FS did not differ from control values. In summary, the expression of CP-AMPARs in entorhinal cortex and hippocampal pyramidal neurons in young rats decreases significantly with FS. These alterations do not impact the characteristics of basal synaptic transmission in the hippocampus.

A novel deep learning model for a coronary computed tomography angiography diagnosis of plaque erosion

Abstract Background Plaque erosion is the underlying pathology in up to 50% of patients with non-ST-segment elevation acute coronary syndromes (ACS). Previous studies suggest that patients with erosion might be treated conservatively without stenting. Currently, a diagnosis of plaque erosion requires invasive intracoronary imaging. Purpose We sought to develop a deep learning (DL) model that enables an accurate diagnosis of plaque erosion using coronary computed tomography angiography (CTA). Methods A total of 532 coronary CTA scans (143 coronary CTA scans with plaque rupture, 113 with plaque erosion, and 276 with reference segment) from 395 patients were included to develop a model: 426 coronary CTA scans from 316 patients for training and internal validation, and 106 separate scans from 79 patients for test set validation. Among ACS patients, the diagnosis of plaque rupture or erosion identified by OCT was used as the ground truth, and the corresponding site on the CTA image matched with OCT image was determined to be the culprit lesion. A novel DL, Momentum Distillation Composite Transformer Attention (MD-CTA) model, was developed that can effectively process the whole CTA scans to diagnose plaque erosion. Results The novel DL model showed diagnostic performance in the five-fold cross-validation with an area under the receiver-operating characteristic curve (AUC) of 0.901 (0.873-0.930), a sensitivity of 81.2 (72.8-88.0), and a specificity of 86.6 (82.4-90.2), all of which were significantly higher than those of convolutional neural network (CNN) model with an AUC of 0.621 (0.567-0.675), the sensitivity of 59.8 (50.1-69.0), and specificity of 60.2 (54.5-65.7) in training validation. Similarly, in the test set validation, the AUC, sensitivity, and specificity of the MD-CTA model were 0.899 (0.841-0.957), 87.1 (70.2-96.4), and 85.3 (75.3-92.4), respectively, higher than those of 0.724 (0.622-0.826), 71.0 (52.0-85.8), and 68.0 (56.2-78.3) of the CNN model. At the slice level, the MD-CTA model was also superior in the five-fold cross-validation and the test set validation compared to the CCN model. This model significantly outperformed experienced cardiologists, especially for sensitivity. Conclusions The newly developed DL model enables an accurate CTA diagnosis of plaque erosion, which might enable the cardiologist to provide tailored therapy without invasive procedures. This algorithm might lead to a major shift in managing millions of patients with plaque erosion each year.Table 1Figure 1

COV-CT-NET - A DEEP LEARNING MODEL FOR COVID-19, COMMUNITY-ACQUIRED PNEUMONIA DETECTION USING CT IMAGES

The world has witnessed the deadly impact of the Novel Corona Virus (COVID-19), claiming millions of lives since its outbreak in early December 2019. Early virus detection plays a crucial role in controlling this highly contagious disease. Though Reverse Transcription Polymerase Chain Reaction (RT-PCR) is the current standard for confirmation of COVID-19, it is time-consuming. Computed Tomography (CT) imaging of the lungs can preferably be used for fast diagnosis of the disease as it is more sensitive and can detect complications. Due to the unavailability of adequate expertise, a deep learning-based model on CT images is a potential solution for fast detecting SARS Cov2 virus. In this study, we developed a simple but robust Convolution Neural Network model with multiclass detection ability between normal lungs, COVID-19 infected lungs and any other Community-Acquired Pneumonia (CAP) infection using Chest CT images. It is tested on a publicly available dataset, COVID-CT-MD and it achieved slice level accuracy of 99% on test dataset. We also attempted slice-level prediction of the unlabelled slices available in the dataset of COVID-19 and CAP cases.

Combined use of body condition score, radiography, ultrasonography and computed tomography in body condition evaluation of crab-eating fox (Cerdocyon thous).

This study aimed to evaluate the application of radiography, ultrasonography, and computed tomography (CT) as auxiliaries in the determination of body fat and musculature in crab-eating foxes (Cerdocyon thous). Eight animals were evaluated and classified based on the body condition score (BCS) scale. The total cut volume, total visceral volume, total subcutaneous volume, visceral fat volume, subcutaneous fat volume, and total fat volume were measured. The correlation of the BCS with the other data was determined. In the ultrasound, there was a correlation of the vertebral epaxial musculature score (VEMS) with body weight, BCS, muscle mass score (MMS), animal's height, and height at the withers. In the radiographic analysis, the amount of subcutaneous fat obtained in L3 vertebrae did not present a significant correlation with BCS, as opposed to L6. In the tomographic analysis, TFV/L6, VFV/L6, and SFV/L6 correlated with BCS in L3, and only SFV /L6 in L6. When the ratios between fat volumes were correlated, the correlation with BCS occurred only in L6 in the TFV/TBV, SFV/SVC, and VFW variants. Radiography and CT were effective and had a good association with BCS. With the CT it was possible to demonstrate the distribution of total fat at the level of the L3 slice. The ratio of calculations considering the length of the L6 vertebra was more adequate, while at the level of L6 the ratio between the volumes of the compartments was more effective.

Seeking an optimal approach for Computer-aided Diagnosis of Pulmonary Embolism

Pulmonary Embolism (PE) represents a thrombus (“blood clot”), usually originating from a lower extremity vein, that travels to the blood vessels in the lung, causing vascular obstruction and in some patients death. This disorder is commonly diagnosed using Computed Tomography Pulmonary Angiography (CTPA). Deep learning holds great promise for the Computer-aided Diagnosis (CAD) of PE. However, numerous deep learning methods, such as Convolutional Neural Networks (CNN) and Transformer-based models, exist for a given task, causing great confusion regarding the development of CAD systems for PE. To address this confusion, we present a comprehensive analysis of competing deep learning methods applicable to PE diagnosis based on four datasets. First, we use the RSNA PE dataset, which includes (weak) slice-level and exam-level labels, for PE classification and diagnosis, respectively. At the slice level, we compare CNNs with the Vision Transformer (ViT) and the Swin Transformer. We also investigate the impact of self-supervised versus (fully) supervised ImageNet pre-training, and transfer learning over training models from scratch. Additionally, at the exam level, we compare sequence model learning with our proposed transformer-based architecture, Embedding-based ViT (E-ViT). For the second and third datasets, we utilize the CAD-PE Challenge Dataset and Ferdowsi University of Mashad’s PE Dataset, where we convert (strong) clot-level masks into slice-level annotations to evaluate the optimal CNN model for slice-level PE classification. Finally, we use our in-house PE-CAD dataset, which contains (strong) clot-level masks. Here, we investigate the impact of our vessel-oriented image representations and self-supervised pre-training on PE false positive reduction at the clot level across image dimensions (2D, 2.5D, and 3D). Our experiments show that (1) transfer learning boosts performance despite differences between photographic images and CTPA scans; (2) self-supervised pre-training can surpass (fully) supervised pre-training; (3) transformer-based models demonstrate comparable performance but slower convergence compared with CNNs for slice-level PE classification; (4) model trained on the RSNA PE dataset demonstrates promising performance when tested on unseen datasets for slice-level PE classification; (5) our E-ViT framework excels in handling variable numbers of slices and outperforms sequence model learning for exam-level diagnosis; and (6) vessel-oriented image representation and self-supervised pre-training both enhance performance for PE false positive reduction across image dimensions. Our optimal approach surpasses state-of-the-art results on the RSNA PE dataset, enhancing AUC by 0.62% (slice-level) and 2.22% (exam-level). On our in-house PE-CAD dataset, 3D vessel-oriented images improve performance from 80.07% to 91.35%, a remarkable 11% gain. Codes are available at GitHub.com/JLiangLab/CAD_PE.

Discrimination between leucine-rich glioma-inactivated 1 antibody encephalitis and gamma-aminobutyric acid B receptor antibody encephalitis based on ResNet18.

This study aims to discriminate between leucine-rich glioma-inactivated 1 (LGI1) antibody encephalitis and gamma-aminobutyric acid B (GABAB) receptor antibody encephalitis using a convolutional neural network (CNN) model. A total of 81 patients were recruited for this study. ResNet18, VGG16, and ResNet50 were trained and tested separately using 3828 positron emission tomography image slices that contained the medial temporal lobe (MTL) or basal ganglia (BG). Leave-one-out cross-validation at the patient level was used to evaluate the CNN models. The receiver operating characteristic (ROC) curve and the area under the ROC curve (AUC) were generated to evaluate the CNN models. Based on the prediction results at slice level, a decision strategy was employed to evaluate the CNN models' performance at patient level. The ResNet18 model achieved the best performance at the slice (AUC = 0.86, accuracy = 80.28%) and patient levels (AUC = 0.98, accuracy = 96.30%). Specifically, at the slice level, 73.28% (1445/1972) of image slices with GABAB receptor antibody encephalitis and 87.72% (1628/1856) of image slices with LGI1 antibody encephalitis were accurately detected. At the patient level, 94.12% (16/17) of patients with GABAB receptor antibody encephalitis and 96.88% (62/64) of patients with LGI1 antibody encephalitis were accurately detected. Heatmaps of the image slices extracted using gradient-weighted class activation mapping indicated that the model focused on the MTL and BG for classification. In general, the ResNet18 model is a potential approach for discriminating between LGI1 and GABAB receptor antibody encephalitis. Metabolism in the MTL and BG is important for discriminating between these two encephalitis subtypes.

Radiomics-Based Quality Control System for Automatic Cardiac Segmentation: A Feasibility Study.

In the past decade, there has been a rapid increase in the development of automatic cardiac segmentation methods. However, the automatic quality control (QC) of these segmentation methods has received less attention. This study aims to address this gap by developing an automatic pipeline that incorporates DL-based cardiac segmentation and radiomics-based quality control. In the DL-based localization and segmentation part, the entire heart was first located and cropped. Then, the cropped images were further utilized for the segmentation of the right ventricle cavity (RVC), myocardium (MYO), and left ventricle cavity (LVC). As for the radiomics-based QC part, a training radiomics dataset was created with segmentation tasks of various quality. This dataset was used for feature extraction, selection, and QC model development. The model performance was then evaluated using both internal and external testing datasets. In the internal testing dataset, the segmentation model demonstrated a great performance with a dice similarity coefficient (DSC) of 0.954 for whole heart segmentations. Images were then appropriately cropped to 160 × 160 pixels. The models also performed well for cardiac substructure segmentations. The DSC values were 0.863, 0.872, and 0.940 for RVC, MYO, and LVC for 2D masks and 0.928, 0.886, and 0.962 for RVC, MYO, and LVC for 3D masks with an attention-UNet. After feature selection with the radiomics dataset, we developed a series of models to predict the automatic segmentation quality and its DSC value for the RVC, MYO, and LVC structures. The mean absolute values for our best prediction models were 0.060, 0.032, and 0.021 for 2D segmentations and 0.027, 0.017, and 0.011 for 3D segmentations, respectively. Additionally, the radiomics-based classification models demonstrated a high negative detection rate of >0.85 in all 2D groups. In the external dataset, models showed similar results. We developed a pipeline including cardiac substructure segmentation and QC at both the slice (2D) and subject (3D) levels. Our results demonstrate that the radiomics method possesses great potential for the automatic QC of cardiac segmentation.

Deep learning-based automated COVID-19 classification from computed tomography images

ABSTRACT This paper introduces a lightweight Convolutional Neural Networks (CNN) method for image classification in COVID-19 diagnosis. The proposed approach emphasizes simplicity while achieving high performance, and it leverages a meticulously annotated database. The CNN model consists of four convolutional layers, followed by flattening and two dense layers. The methodology focuses on classifying 2D slices of Computed Tomography (CT) scans. To enhance accuracy, the slices undergo anatomy-relevant masking and the removal of non-representative slices from the CT volume. This is achieved by cropping a fixed-sized rectangular area to capture the relevant region of interest and using a threshold based on bright pixels in binarized slices. The proposed methodology demonstrates improved quantitative results in slice classification by employing slice processing techniques. Additionally, augmentation techniques such as class weight balancing, slice flipping, and a learning rate scheduler are applied to diagnose at the slice level. For patient-level diagnosis, a majority voting method is employed by considering the slices of each CT scan. The proposed method surpasses the baseline approach and other alternatives in terms of macro F1 score, both on the validation set and a test partition containing previously unseen images from the rigorously annotated dataset.

Siam-VAE: A hybrid deep learning based anomaly detection framework for automated quality control of head CT scans.

An automated quality control (QC) system is essential to ensure streamlined head computed tomography (CT) scan interpretations that do not affect subsequent image analysis. Such a system is advantageous compared to current human QC protocols, which are subjective and time-consuming. In this work, we aim to develop a deep learning-based framework to classify a scan to be of usable or unusable quality. Supervised deep learning models have been highly effective in classification tasks, but they are highly complex and require large, annotated data for effective training. Additional challenges with QC datasets include - 1) class-imbalance - usable cases far exceed the unusable ones and 2) weak-labels - scan level labels may not match slice level labels. The proposed framework utilizes these weak labels to augment a standard anomaly detection technique. Specifically, we proposed a hybrid model that consists of a variational autoencoder (VAE) and a Siamese Neural Network (SNN). While the VAE is trained to learn how usable scans appear and reconstruct an input scan, the SNN compares how similar this input scan is to its reconstruction and flags the ones that are less similar than a threshold. The proposed method is more suited to capture the differences in non-linear feature structure between the two classes of data than typical anomaly detection methods that depend on intensity-based metrics like root mean square error (RMSE). Comparison with state-of-the-art anomaly detection methods using multiple classification metrics establishes superiority of the proposed framework in flagging inferior quality scans for review by radiologists, thus reducing their workload and establishing a reliable and consistent dataflow.

Automatic deep learning method for detection and classification of breast lesions in dynamic contrast-enhanced magnetic resonance imaging.

The purpose of this study was to develop a deep learning-based system with a cascade feature pyramid network for the detection and classification of breast lesions in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). This retrospective study enrolled 191 consecutive patients with pathological confirmed breast lesions who underwent preoperative magnetic resonance imaging (MRI) at the Second Affiliated Hospital of Xi'an Jiaotong University. Patients were randomly divided into a training set comprising 153 patients with 126 malignant and 27 benign lesions and a validation set containing 38 patients with 31 malignant and 7 benign lesions under 5-fold cross-validation. Two radiologists annotated the location and classification of all lesions. After augmentation with pseudo-color image fusion, MRI images were fed into the developed cascade feature pyramid network system, feature pyramid network, and faster region-based convolutional neural network (CNN) for breast lesion detection and classification, respectively. The performance on large (>2 cm) and small (≤2 cm) breast lesions was further evaluated. Average precision (AP), mean AP, F1-score, sensitivity, and false positives were used to evaluate different systems. Cohen's kappa scores were calculated to assess agreement between different systems, and Student's t-test and the Holm-Bonferroni method were used to compare multiple groups. The cascade feature pyramid network system outperformed the other systems with a mean AP and highest sensitivity of 0.826±0.051 and 0.970±0.014 (at 0.375 false positives), respectively. The F1-score of the cascade feature pyramid network in real detection was also superior to that of the other systems at both the slice and patient levels. The mean AP values of the cascade feature pyramid network reached 0.779±0.152 and 0.790±0.080 in detecting large and small lesions, respectively. Especially for small lesions, the cascade feature pyramid network achieved the best performance in detecting benign and malignant breast lesions at both the slice and patient levels. The deep learning-based system with the developed cascade feature pyramid network has the potential to detect and classify breast lesions on DCE-MRI, especially small lesions.

Efficient labelling for efficient deep learning: the benefit of a multiple-image-ranking method to generate high volume training data applied to ventricular slice level classification in cardiac MRI.

Getting the most value from expert clinicians' limited labelling time is a major challenge for artificial intelligence (AI) development in clinical imaging. We present a novel method for ground-truth labelling of cardiac magnetic resonance imaging (CMR) image data by leveraging multiple clinician experts ranking multiple images on a single ordinal axis, rather than manual labelling of one image at a time. We apply this strategy to train a deep learning (DL) model to classify the anatomical position of CMR images. This allows the automated removal of slices that do not contain the left ventricular (LV) myocardium. Anonymised LV short-axis slices from 300 random scans (3,552 individual images) were extracted. Each image's anatomical position relative to the LV was labelled using two different strategies performed for 5 hours each: (I) 'one-image-at-a-time': each image labelled according to its position: 'too basal', 'LV', or 'too apical' individually by one of three experts; and (II) 'multiple-image-ranking': three independent experts ordered slices according to their relative position from 'most-basal' to 'most apical' in batches of eight until each image had been viewed at least 3 times. Two convolutional neural networks were trained for a three-way classification task (each model using data from one labelling strategy). The models' performance was evaluated by accuracy, F1-score, and area under the receiver operating characteristics curve (ROC AUC). After excluding images with artefact, 3,323 images were labelled by both strategies. The model trained using labels from the 'multiple-image-ranking strategy' performed better than the model using the 'one-image-at-a-time' labelling strategy (accuracy 86% vs. 72%, P=0.02; F1-score 0.86 vs. 0.75; ROC AUC 0.95 vs. 0.86). For expert clinicians performing this task manually the intra-observer variability was low (Cohen's κ=0.90), but the inter-observer variability was higher (Cohen's κ=0.77). We present proof of concept that, given the same clinician labelling effort, comparing multiple images side-by-side using a 'multiple-image-ranking' strategy achieves ground truth labels for DL more accurately than by classifying images individually. We demonstrate a potential clinical application: the automatic removal of unrequired CMR images. This leads to increased efficiency by focussing human and machine attention on images which are needed to answer clinical questions.