Automated Research Articles

Automatic and Rapid Throughflow-Based Optimal Design Method for Multi-Stage Axial-Flow Compressors

Automatic and Rapid Throughflow-Based Optimal Design Method for Multi-Stage Axial-Flow Compressors

Medical Image Analysis Methods for Accurate Segmentation and Quantification of Acute-Subacute Ischemic Stroke Lesion — A Systematic Review

A medical emergency, acute ischemic stroke necessitates immediate treatment as the longer the neural is without blood flow, the greater the risk of irreversible neural damage and disability. Depending on the extent, the lesion may vary in size and location and location of the blood flow blockage, and can have significant effects on a person’s cognitive, motor, and sensory function. The process of identifying and classifying regions of affected neural tissue during a ischemic stroke is referred to as “ischemic stroke lesion segmentation.” Medical imaging techniques like computed tomography (CT) and magnetic resonance imaging (MRI) are typically used to accomplish this specialized software tools. The goal of segmentation is to provide a detailed and accurate map of the extent and location of the stroke damage, which can be used to guide treatment decisions and monitor the patient’s recovery. Although manual segmentation is regarded as the most accurate method, it takes time and is subject to variability between observers, which leads to inconsistent results. There are a few unique ways to deal with dividing ischemic stroke sores, ranging from manual delineation by trained clinician to fully automated algorithms that use machine learning and other advanced techniques. Automated segmentation methods are faster and more objective, but they may require large training sets and can be sensitive to imaging artifacts and other sources of noise. Although fully automatic methods hold promise, semi-automatic methods remain the preferred approach in clinical research. We performed a systematic review of the literature to explore the latest advancements and trends in analyzing ischemic stroke lesions using automated methods developed in the past five years. Our search of IEEE explores, Springer, Science Direct, Taylor & Francis, etc. yielded 1580 papers, from which we selected 50 for detailed analysis. Of these studies, 12 employed supervised segmentation, 12 used unsupervised segmentation and 27 employed deep learning segmentation methods. Only a limited number of studies have validated their fully automatic methods using longitudinal samples, and a mere eight studies included validation using clinical parameters. Furthermore, only 23 of the 50 studies made their methods publicly available. To advance the field, there is a need for fully automatic methods that are validated with longitudinal samples and clinical parameters. Moreover, making methods publicly available is essential for promoting reproducibility and facilitating comparison of results.

Understanding the stress field at the lateral termination of a thrust fold using generic geomechanical models and clustering methods

Abstract. This study employs numerical simulations based on the limit analysis (LA) method to calculate the stress distribution in a model that includes a basal detachment, featuring the lateral termination of a generic fault under compression. We conduct 2500 2D and 500 3D simulations with varying basement and fault friction angles to analyze and classify the results into clusters representing similar failure patterns to understand the stress fields. Automatic fault detection methods are employed to identify the number and positions of fault lines in 2D and fault surfaces in 3D. Clustering approaches are utilized to group the models based on the detected failure patterns. For the 2D models, the analysis reveals three primary clusters and five transitional ones, qualitatively consistent with the critical Coulomb wedge theory and the influence of inherited structural and geometric aspects over rupture localization. In the 3D models, four different clusters portray the lateral prolongation of the inherited fault. High stress magnitudes are detected between the compressive boundary and the activated or created faults and at the root of the inherited active fault. Tension zones appear near the outcropping surface relief, while stress decreases with depth at the footwall of the created back thrusts. A statistical cluster-based stress field analysis indicates that for a given cluster, the stress field mainly conserves the same orientations, while the magnitude varies with changes in friction angles and compressive field intensity, except in failure zones where variations are sparse. Small parametric variations could lead to significantly different stress fields, while larger deviations might result in similar configurations. The comparison between 2D and 3D models shows the importance of lateral stresses and their influence on rupture patterns, distinguishing between 3D analysis and 2D cross-sections. Lastly, despite using small-scale models, stress field variations over a span of a couple of kilometers are quite large.

Methods for the Construction and Editing of an Efficient Control Network for the Photogrammetric Processing of Massive Planetary Remote Sensing Images

Planetary photogrammetry remains an important technical means of producing high-precision planetary maps. High-quality control networks are fundamental to successful bundle adjustment. However, current software tools used by the planetary mapping community to construct and edit control networks exhibit very low efficiency. Moreover, redundant and invalid control points in the control network can further increase the time required for the bundle adjustment process. Due to a lack of targeted algorithm optimization, existing software tools and methods are unable to meet the photogrammetric processing requirements of massive planetary remote sensing images. To address these issues, we first proposed an efficient control network construction framework based on approximate orthoimage matching and hash quick search. Next, to effectively reduce the redundant control points in the control network and decrease the computation time required for bundle adjustment, we then proposed a control network-thinning algorithm based on a K-D tree fast search. Finally, we developed an automatic detection method based on ray tracing for identifying invalid control points in the control network. To validate the proposed methods, we conducted photogrammetric processing experiments using both the Lunar Reconnaissance Orbiter (LRO) narrow-angle camera (NAC) images and the Origins Spectral Interpretation Resource Identification Security Regolith Explorer (OSIRIS-REx) PolyCam images; we then compared the results with those derived from the famous open-source planetary photogrammetric software, the United States Geological Survey (USGS) Integrated Software for Imagers and Spectrometers (ISIS) version 8.0.0. The experimental results demonstrate that the proposed methods significantly improve the efficiency and quality of constructing control networks for large-scale planetary images. For thousands of planetary images, we were able to speed up the generation and editing of the control network by more than two orders of magnitude.

Molecular characterization based on cytochrome C oxidase I gene of the family Channidae from different riverine systems of Odisha, India

This study focuses on four Channa species from the northern part of the Eastern Ghats in India: Channa punctata (Bloch, 1793); Channa striata (Bloch, 1793); Channa marulius (Hamilton, 1822) and Channa gachua (Hamilton, 1822). The aim is to determine whether their molecular divergence aligns with their morphological distinctiveness. The molecular analysis based on mitochondrial COI gene sequences revealed distinct clusters for each species and genus, with K2P distances ranging from 17.28 – 27.96%. Notably, C. marulius was positioned in a distinct clade separate from the C. punctata, C. striata, and C. gachua groups. Channa gachua clustered with Channa rara and Channa kelaartii, suggesting a close evolutionary relationship. Channa punctata and C. striata also clustered together, reinforcing their status as sister species. Species delimitation using the Assemble Species by Automatic Partitioning method identified 10 partitions, with the most reliable delimitation showing a clear separation of Channa species of Odisha. These findings align with recent analyses and confirm that the molecular divergence among the four species is consistent with their morphological differences. The study underscores the importance of molecular methods in resolving taxonomic ambiguities and understanding species diversity. It provides a foundational molecular database for Channa species, supporting future research on genetic divergence and contributing to conservation and aquaculture efforts.

A Comparison of Manual and Automatic Methods for Calculating Cumulative Dose-Volume Histogram Parameters in Cervical Cancer Radiotherapy

Background: The cumulative dose-volume histogram (DVH) parameters for organs at risk (OAR) and the target volume, obtained from external beam radiotherapy (EBRT) and brachytherapy (BT), are commonly calculated manually in clinical practice. Objectives: This study aims at comparing the manual and automatic methods for calculating cumulative DVH parameters of OAR and the target volume in patients with cervical cancer. Methods: The cumulative DVH of EBRT and BT methods for target volume and OAR, including rectum, bladder, and sigmoid was calculated manually, using the EQD2 (equivalent dose in 2 Gy fractions) formula. In the automatic method, dosimetric data were collected as DICOM files and entered into SagiPlan® BT software. Using the BED/EQD2 summation feature, the BT and EBRT plans were summed, and the cumulative DVH parameters were extracted. The results were compared with the manual method. Results: In 32 cervical cancer samples, dosimetric comparisons showed significant differences between manual (EQD2) and automatic (SagiPlan®) calculations for rectum (mean dose 69.15 vs. 71.92 Gy, P < 0.01), bladder (mean dose 81.87 vs. 84.47 Gy, P < 0.01), and sigmoid (mean dose 69.84 vs.73.15 Gy, P < 0.01). However, the cumulative dose to the target tissue was similar between the two methods (mean dose 86.9 vs. 86.91, P = 0.21). Automatic calculations indicated higher doses of the OARs, suggesting potential underestimation by manual methods. Conclusions: In clinical practice, when designing BT planning for cervical cancer, it is recommended to input the patient’s EBRT design data into the planning software to determine the cumulative dose received by the target and OARs.

NasalSeg: A Dataset for Automatic Segmentation of Nasal Cavity and Paranasal Sinuses from 3D CT Images

Modern facial surgical planning and therapeutic strategies rely heavily on the precise segmentation of the nasal cavity and paranasal sinuses from computed tomography (CT) images for quantitative analysis. Nevertheless, manual segmentation is labor-intensive and prone to inconsistencies, highlighting the need for automatic segmentation methods. A significant challenge in this field is the lack of publicly available clinical datasets for research. To address this issue, we introduce NagalSeg, the first large-scale, publicly available dataset for nasal cavity and paranasal sinus segmentation. In comparison to existing nasal structure segmentation datasets, which are either private or small-scale, NagalSeg stands out as the first publicly accessible dataset. It provides an order of magnitude more labeled data, consisting of 130 3D CT scans with pixel-wise annotations of five anatomical structures: the left nasal cavity, right nasal cavity, nasopharynx, left maxillary sinus, and right maxillary sinus. The NagalSeg dataset serves as an open-access resource to facilitate the development and evaluation of segmentation algorithms and promote future in-depth research towards the clinical application of artificial intelligence methods.

A Pilot Study of PSMA PET/CT and MRI Fusion for Prostate Cancer: Software to Replace PET/MRI Hardware.

Introduction: Prostate-specific membrane antigen positron emission tomography/computed tomography (PSMA PET/CT), in combination with magnetic resonance imaging (MRI), may enhance the diagnosis and staging of prostate cancer. Image fusion of separately acquired PET/CT and MRI images serve to facilitate clinical integration and treatment planning. This study aimed to investigate different PSMA PET/CT and MRI image fusion workflows for prostate cancer visualisation. Methods: Eighteen patients with prostate cancer who underwent PSMA PET/CT and MRI prior to radical prostatectomy were retrospectively selected. Alignment of the prostate was performed between PET/CT and MRI via three techniques: semi-automatic rigid, automatic rigid, and automatic non-rigid. Image fusion accuracy was evaluated through boundary and volume agreement, quantified by the Dice Similarity Coefficient (DSC), 95% Hausdorff Distance (HD), and Mean Surface Distance (MSD), with comparison against reconstructed histopathology slices. Results: Image fusion using all techniques resulted in clear lesion visualisation from PSMA PET/CT overlay and anatomical detail afforded by the MRI base and was consistent with histopathology tumour location. Image fusion accuracy was within the recommended range based on a DSC of 0.8-0.9. The automatic non-rigid registration method had the highest volume agreement (DSC: 0.96 ± <0.01) and boundary agreement (HD: 1.17 ± 0.35 mm) when compared to automatic rigid (DSC 0.88 ± 0.02, HD 3.18 ± 0.29 mm) and semi-automatic rigid (DSC 0.80 ± 0.06, HD 5.25 ± 1.68 mm). Conclusions: Image fusion of clinically obtained PET/CT and MRI is feasible and clinically acceptable for use in prostate cancer diagnosis and surgical management. While the best accuracy was observed with the automatic non-rigid technique, which requires further validation, image fusion with clinically accessible methods (semi-automatic rigid) may currently aid patient education, pre-operative planning, and intra-operative guidance.

Multi-parameter coupling modeling method and hybrid mechanism of concrete-encased CFST hybrid structures

Concrete-encased concrete-filled steel tubular (CFST) hybrid structures consist of encased CFST components and reinforced concrete (RC) encasement, posing unique challenges related to their hybrid mechanisms and coupling effects. This study addresses these issues by proposing an automatic and efficient Finite Element (FE) modeling method, which accounts for multiple constitutive models of confined concrete, meticulous grid division and reasonable interaction. The FE modeling method was packaged with a user-friendly graphical interface in the software ‘Auto CECFST’, which has been shared on GitHub(https://github.com/CECFST/Auto-CECFST.git). The modeling method has been verified by test results on the strength, stiffness and deformation capacity. Utilizing the refined FE modeling method, 7776 FE models are generated based on an orthogonal combination of 7 critical parameters of concrete-encased CFST hybrid structures. To accurately define failure modes, stress ratio (φ) was proposed to achieve quantitative analysis on the failure mode, followed by a comprehensive investigation into the full-range analysis of the hybrid mechanism of two components. Moreover, further exploration focused on multi-parameter coupling effects of critical mechanism characteristics, including strength, stiffness, and deformation capacity, elucidating the hybrid mechanism and mechanical similarities between single and multi-chord structures. Based on the above analysis, the balance of strength and deformation ability could be achieved by certain parameter ranges. Finally, available strength and stiffness calculation methods are validated against experimental and FE results under three typical failure modes, revealing the advantages and limitations of calculation methods on the basis of mechanics and data statistics, which provides a basis for security structural design.

A Systematic Literature Review on Automatic Sexism Detection in Social Media

Sexist content has become increasingly prevalent on social media platforms, underscoring the critical need for the development of efficient Automatic Sexism Detection methods. Previous literature reviews have not encompassed the new advancements in Automatic Sexism Detection observed over the past three years. Hence, the present study conducted a Systematic Literature Review (SLR) that examined 48 primary studies published between 2014 and 17th Sept. 2024, retrieved from six bibliographic databases. This paper aims to present a comprehensive literature review on Automatic Sexism Detection, encompassing the datasets, preprocessing techniques, feature extraction methods, text representations, classification approaches, and evaluation models employed in Automatic Sexism Detection research. The paper includes a discussion of the findings, limitations, and future research directions of the chosen articles. Additionally, it provides an overview of the conclusions drawn from the conducted research. The performed analysis reveals a lack of corpus beyond the English and Spanish language encountered in datasets, with most of the latter being annotated for either misogyny or non-misogyny. Common preprocessing techniques analyzed in the current study include lowercase conversion, text removal, tokenization, stemming, and rewriting. Discrete representations, such as TF-IDF, N-grams, and BoW, are frequently utilized, while distributed representations, like Bert and GloVe, are prominent. Bert is the predominant classification model utilized while combining lexical features can enhance the results in the majority of the discussed scenarios. Accuracy (A) and F1 score (F1) are the most widely deployed evaluation metrics in this field.

Superpixel-based principal feature clustering annotation method for dual-phase microstructure segmentation

Metallographic analysis is one of the most commonly used techniques by materials scientists for studying metal materials. The deep learning methods, which have been widely applied in metallographic images analysis, demonstrate excellent performance in this task. However, the optimization of deep learning models often relies on a substantial amount of accurately labeled samples for effective supervision. To address this issue, this paper proposes a novel automatic annotation method based on brightness and spatial distribution which is suitable for deep learning-based segmentation of optical dual-phase metallographic microstructure. The proposed automatic annotated method includes the superpixel segmentation, principal feature extraction, and clustering algorithm therefore it is referred as the superpixel-based principal feature clustering annotation (SPFCA) method. SPFCA employs discriminative criteria similar to those used by metallurgists to differentiate between metallographic structures. Furthermore, it can mitigate the occasional errors inherent in manual annotation, leading to improved performance compared to models trained with expert annotations. Experimental validation was conducted using four self-built datasets with different image qualities to test the performance of models from different perspective. Initially, hyperparameter optimization for the SPFCA method tailored to our dataset was performed. Subsequently, SPFCA was utilized to guide the optimization of the convolutional neural network employed for segmentation. The results demonstrate that the segmentation model optimized with SPFCA guidance achieved an F1 score of 0.9226 in the single dataset without the need for manual labeling, surpassing the segmentation models optimized with expert annotations.

Palm Oil Tree Canopy Identification Using Deep Learning Approach (Case Study: Tanjung Gusta District, North Sumatera)

Abstract The palm oil plantation industry in Indonesia has growing rapidly as demand for palm oil increases globally. This needs to be supported by technological innovation to increase palm oil production. One of them is to integrate the power of artificial intelligence technology. This research aims to develop a robust and accurate method for segmenting oil palm trees in plantation areas. Leveraging deep learning algorithms techniques, the research explores the potential of SAM in accurately delineating individual oil palm trees derived from aerial imagery data. The study also involves the development of a comprehensive and versatile labelled dataset to support the training and validation of the deep learning models for oil palm tree counting and segmentation. The performance of the proposed approach is evaluated and discussed critically. This research demonstrates the potential of deep learning algorithms for large-scale mapping and accurate counting of oil palm trees in plantation areas. The author hopes that the result and analysis of this research will give insight and improvement in detecting oil palm trees using automatic method.

Building Footprint Extraction from Fixed-Wing UAV Imagery using Mask R-CNN and Object-based Image Analysis Methods (Case Study: Banturejo Village, Malang Regency)

Abstract As a developing area in Malang Regency, Banturejo Village has many potencies since its location near the tourism area of Selorejo Dam. To maximally the harness of potencies while maintaining efficient land use in Banturejo village, mapping the built area in large scale should be carried out. The photogrammetric techniques using fixed-wing UAV could be a good alternative for large-scale mapping in this village area because of its capability to quickly acquire high resolution image with highly customizable mission specifications. But the problem arises in interpreting these imagery into meaningful cartographic representation which often requires cautious manual digitization in much slower rate that its acquisition. In this research the automatic image analysis method for building footprint extraction using Mask R-CNN algorithm and Object-Based Image Analysis was performed. The fixed wing UAV imagery was captured in 2023 and the structure from motion algorithm was employed for photogrammetric processing which produced 10-cm resolution orthophoto. Manually digitized building polygons from the same imagery serve as the gold standard for accuracy analysis, and small proportion of the data was used as training samples for the algorithm. The results shows that 1447 buildings with total area of 180,595 m2 was generated with Mask R-CNN algorithm, while OBIA-Mask R-CNN produced 572 buildings and total area of 201,932 m2. The confusion matrices reveal precision value of 77.94%, recall 51.54%, F1 Score 62.02% by Mask R-CNN method, and precision value of 35.95%, recall 9.21%, F1 Score 14.66% by OBIA-Mask RCNN method. Mask R-CNN method generated slightly lower accuracy of total building area, but in terms of precision the OBIA-Mask RCNN method produces lower number of building polygons.

Handling non-linearities and pre-processing in multivariate calibration of vibrational spectra

Backgroundseveral automatic pre-processing methods are available for coping with scattering effects in vibrational spectra (near and mid infrared, Raman) when linear multivariate models are applied for calibration. In contrast, almost no methods have been developed for data sets showing a non-linear relationship between instrumental signals and analyte concentrations or target properties. Resultsseveral data sets are studied, both simulated and experimental, showing different degrees of non-linearity: very low, intermediate, and high. For each of them, three multivariate models have been applied: classical partial least-squares (PLS), sequential pre-processing through orthogonalization (SO-PLS) and Kernel partial least-squares (K-PLS). The success of these models depends on the degree of non-linearity presented by each data set. The results are discussed in terms of the model structures. Significancethe need of developing new automatic pre-processing techniques for non-linear vibrational data sets is highlighted, in order to reach a similar status as the linear counterparts.

Flood Inundation Distribution In Lamongan Regency Using Sentinel-1 SAR Data And Automatic Water-S1 Method

Abstract The study of SAR data in mapping distribution flood inundation has yielded relevant results. The ease of obtaining data has made development application of Sentinel-1 SAR, for flood mapping. A common method used in flood mapping is thresholding, mainly by determining the threshold generated through visual observation that presented color contrast histograms. However, surface conditions of surrounding terrain can create misinformation due to increased surface roughness, like double bounce on vegetation make it more challenging. Moreover, manual threshold methods are less-time effective in trial error process from visual image interpretation. The results tend to be subjective because distribution mapping heavily relies on the operator’s visual interpretation of threshold value. To address this challenge, distribution of flood inundation will be mapped with an automatic threshold method, called Water-S1 and expected to provide more objective results. The images divided into tiles with 10 km size, each selected tiles will be modeled with data probability parameters. Distribution modeling then iterated get probability model distribution that represents each selected tile. The result is probability map of the water surface during and after flood, then subtracts to get the distribution area of flood inundation. In this research the method was applied to Lamongan Regency. The result show that the method can identify flood inundation quite satisfactory. The spatial distribution of flood inundation are detected clustered in paddy fields center of Lamongan Regency with flood area around 19,680.75 hectares.

An Automatic Optimization Method for Electrical and Thermal Performance of Stacked DBC Power Module

An Automatic Optimization Method for Electrical and Thermal Performance of Stacked DBC Power Module

Development of an automatic tubular chemiluminescence immunoassay using magnetic particles for the detection of antibodies against foot-and-mouth disease virus serotype A.

Foot-and-mouth disease virus (FMDV) causes significant financial losses in animal farms worldwide. Specific antibody monitoring of FMDV post-vaccination is important for evaluating vaccine efficacy, as well as the prevention and control of foot-and-mouth disease (FMD). In this study, we developed a fully automatic tubular chemiluminescence immunoassay method based on magnetic particles (A-MPCLIA) to specifically detect antibodies against FMDV serotype A. The diagnostic sensitivity (Dsn) and diagnostic specificity (Dsp) of A-MPCLIA were determined to be 96.05% and 99%, respectively, with a cut-off value of 30 activity units (U) by detecting the known background swine, cattle, and sheep serum. The developed method presented good Dsn, Dsp, and repeatability (CV < 10%). The activity units of A-MPCLIA, determined using 23 animals vaccinated with FMDV serotypes O, A, and Asia 1 univalent inactivated vaccine and nine serum samples from naïve animals, were significantly positively correlated with the titers of liquid-phase-blocking enzyme-linked immunosorbent assay (R2 = 0.992) and the titer of the virus neutralization test (R2 = 0.9577). In addition, A-MPCLIA requires less than 20 min, and it is fully automatic in detection. These results showed that A-MPCLIA is a suitable method for the detection of antibodies against FMDV serotype A in the future.

A high-precision automatic diagnosis method of maize developmental stage based on ensemble deep learning with IoT devices

Accurately determining the stage of crop development holds significant importance for field crop management. With the advancement of smart agriculture, an increasing number of Internet of Things (IoT) devices are being integrated into agricultural production, enabling more efficient acquisition of high-precision crop images. Currently, research on detecting crop growth stages based on IoT device images remains relatively scarce. Most existing studies rely on a single network model for detection, often encountering issues such as low accuracy and overfitting. Therefore, in this study, we collected maize images using IoT devices and constructed an integrated deep learning model by utilizing four convolutional neural networks (CNNs) to detect the growth period of maize in real time. Additionally, we implemented several improvements on these four CNNs and subsequently tested the performance of the ensemble model on the maize dataset. Regarding the ensemble strategy for the ensemble model, we proposed a dynamic weighted voting method, building upon the original voting approach, which can mitigate model training fluctuations and expedite model convergence. Ultimately, we manually simulated various lighting conditions to assess their impact on the ensemble model. Experimental results demonstrate that the ensemble deep model proposed in this paper represents a robust method for detecting maize growth stages, achieving an accuracy rate of 0.976 on the maize dataset, effectively facilitating high-precision detection of maize growth stages in complex backgrounds.

Analysis of automatic fetal intracranial volume (ICV) measurement based on the optimized ultrasound Smart ICV method at 16-34 weeks of gestation.

Fetal intracranial volume (ICV) can help evaluate the development of the prenatal central nervous system (CNS) from the three-dimensional (3D) attributes of the cranial structure. Accurate and rapid segmentation and calculation of the ICV are clinically significant. Virtual organ computer-aided analysis (VOCAL) is a commonly used method for measuring fetal ICV. However, its operation is highly complex and time-consuming. This study aimed to optimize the fetal Smart ICV method at 16-19 gestational weeks, verify the consistency of automatic and manual measurement of ICV, and assess an automatic and efficient method for evaluating fetal ICV growth in the second and third trimester of pregnancy. The ultrasound data of 950 healthy fetuses at 16-34 weeks of gestation were collected. First, the Smart ICV algorithm was optimized at 16-19 weeks. Second, the optimized Smart ICV was compared with the manual VOCAL method. Finally, growth curve and Z-score estimations for fetuses were established for growth assessment via optimized Smart ICV. Compared with the nonoptimized version, the optimized Smart ICV yielded a lower Hausdorff distance (1.15±0.25 vs. 1.31±0.93 mm, P<0.05). Both intra- and inter-observer agreements were at a high level for ICV measurement based optimized Smart ICV [intra-observer intraclass correlation coefficient (ICC) =0.998, 95% confidence interval (CI): 0.996-0.999; inter-observer ICC =0.991, 95% CI: 0.988-0.996] and the 18 plane-VOCAL (intra-observer ICC =0.997, 95% CI: 0.995-0.998; inter-observer ICC =0.981, 95% CI: 0.979-0.991). Additionally, Bland-Altman analysis showed that the ICV data for the above two models had good agreement. Nevertheless, compared with the 18 plane-VOCAL, the optimized Smart ICV consumed less time (3.7±0.7 vs. 153.1±29.5 s, P<0.05). The best fitting model of gestational week for the Smart ICV was a cubic function, expressed as follows: y = -44.2445 + 0.1427x 2 + 0.0052x 3, where y is ICV and x is the gestational week. In addition, fetal ICV showed an accelerated growth trend in the second trimester. The optimized Smart ICV showed excellent accuracy and efficiency in ICV measurements at 16-34 gestational weeks. Our results may help to establish a best-fit growth curve for ICV. Our findings suggest that the optimized Smart ICV method has the potential to be a reliable tool for fetal growth assessment during the second and third trimesters.

Automatic deep learning method for analysis and prediction of neonatal hyperbilirubinemia in magnetic resonance imaging.

This study conducted a comparative analysis among newborns with varying levels of hyperbilirubinemia, explored the relationships between magnetic resonance imaging (MRI) image features and serum bilirubin levels in hyperbilirubinemia, and proposed an automatic classification system based on deep learning (DL) for prediction of neonatal hyperbilirubinemia (NHB). This retrospective study enrolled 606 consecutive neonates who had their serum bilirubin detected at the Xi'an Fourth Hospital, including 273 cases of patients and 333 cases of normal controls. After data preprocessing, MRI images were fed into the Inception-v3 network, graph convolutional network (GCN), and 3-dimensional (3D) patch-based GCN that introduced the graph attention mechanism (our GCN) for NHB analysis and classification, respectively. Multi-threshold grouping was conducted based on various serum bilirubin levels. Performance evaluation involved the area under the curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE). As the bilirubin levels gradually increased, the overall performance metrics of DL system for detecting the T1-weighted imaging signal in the pallidum region showed a significant upward trend. Our GCN for the prediction and classification of MRI image features of NHB achieved satisfactory results. When the bilirubin value exceeded 400 µmol/L, it achieved an AUC of 0.86 and ACC of 0.81, which is significantly higher than other advanced models (ACC: 72-78.3%) with the same proposed input form. The DL system has the potential to automatically analyze and predict NHB on MRI.