Ground Truth Research Articles

Multi-granularity learning of explicit geometric constraint and contrast for label-efficient medical image segmentation and differentiable clinical function assessment

Automated segmentation is a challenging task in medical image analysis that usually requires a large amount of manually labeled data. However, most current supervised learning based algorithms suffer from insufficient manual annotations, posing a significant difficulty for accurate and robust segmentation. In addition, most current semi-supervised methods lack explicit representations of geometric structure and semantic information, restricting segmentation accuracy. In this work, we propose a hybrid framework to learn polygon vertices, region masks, and their boundaries in a weakly/semi-supervised manner that significantly advances geometric and semantic representations. Firstly, we propose multi-granularity learning of explicit geometric structure constraints via polygon vertices (PolyV) and pixel-wise region (PixelR) segmentation masks in a semi-supervised manner. Secondly, we propose eliminating boundary ambiguity by using an explicit contrastive objective to learn a discriminative feature space of boundary contours at the pixel level with limited annotations. Thirdly, we exploit the task-specific clinical domain knowledge to differentiate the clinical function assessment end-to-end. The ground truth of clinical function assessment, on the other hand, can serve as auxiliary weak supervision for PolyV and PixelR learning. We evaluate the proposed framework on two tasks, including optic disc (OD) and cup (OC) segmentation along with vertical cup-to-disc ratio (vCDR) estimation in fundus images; left ventricle (LV) segmentation at end-diastolic and end-systolic frames along with ejection fraction (LVEF) estimation in two-dimensional echocardiography images. Experiments on nine large-scale datasets of the two tasks under different label settings demonstrate our model’s superior performance on segmentation and clinical function assessment.

Development of a Statistical Model for Automated Ground Truth Generation in Low-Resource Languages

Development of a Statistical Model for Automated Ground Truth Generation in Low-Resource Languages

Learning Ground Displacement Signals Directly from InSAR-Wrapped Interferograms.

Monitoring ground displacements identifies potential geohazard risks early before they cause critical damage. Interferometric synthetic aperture radar (InSAR) is one of the techniques that can monitor these displacements with sub-millimeter accuracy. However, using the InSAR technique is challenging due to the need for high expertise, large data volumes, and other complexities. Accordingly, the development of an automated system to indicate ground displacements directly from the wrapped interferograms and coherence maps could be highly advantageous. Here, we compare different machine learning algorithms to evaluate the feasibility of achieving this objective. The inputs for the implemented machine learning models were pixels selected from the filtered-wrapped interferograms of Sentinel-1, using a coherence threshold. The outputs were the same pixels labeled as fast positive, positive, fast negative, negative, and undefined movements. These labels were assigned based on the velocity values of the measurement points located within the pixels. We used the Parallel Small Baseline Subset service of the European Space Agency's GeoHazards Exploitation Platform to create the necessary interferograms, coherence, and deformation velocity maps. Subsequently, we applied a high-pass filter to the wrapped interferograms to separate the displacement signal from the atmospheric errors. We successfully identified the patterns associated with slow and fast movements by discerning the unique distributions within the matrices representing each movement class. The experiments included three case studies (from Italy, Portugal, and the United States), noted for their high sensitivity to landslides. We found that the Cosine K-nearest neighbor model achieved the best test accuracy. It is important to note that the test sets were not merely hidden parts of the training set within the same region but also included adjacent areas. We further improved the performance with pseudo-labeling, an approach aimed at evaluating the generalizability and robustness of the trained model beyond its immediate training environment. The lowest test accuracy achieved by the implemented algorithm was 80.1%. Furthermore, we used ArcGIS Pro 3.3 to compare the ground truth with the predictions to visualize the results better. The comparison aimed to explore indications of displacements affecting the main roads in the studied area.

A Post-Processing Multipath/NLoS Bias Estimation Method Based on DBSCAN.

Positioning based on Global Navigation Satellite Systems (GNSSs) in urban environments always suffers from multipath and Non-Line-of-Sight (NLoS) effects. In such conditions, the GNSS pseudorange measurements can be affected by biases disrupting the GNSS-based applications. Many efforts have been devoted to detecting and mitigating the effects of multipath/NLoS, but the identification and classification of such events are still challenging. This research proposes a method for the post-processing estimation of pseudorange biases resulting from multipath/NLoS effects. Providing estimated pseudorange biases due to multipath/NLoS effects serves two main purposes. Firstly, machine learning-based techniques can leverage accurately estimated pseudorange biases as training data to detect and mitigate multipath/NLoS effects. Secondly, these accurately estimated pseudorange biases can serve as a benchmark for evaluating the effectiveness of the methods proposed to detect multipath/NLoS effects. The estimation is achieved by extracting the multipath/NLoS biases from pseudoranges using a clustering algorithm named Density-Based Spatial Clustering of Applications with Noise (DBSCAN). The performance is demonstrated using two real-world data collections in multipath/NLoS scenarios for both static and dynamic conditions. Since there is no ground truth for the pseudorange biases due to the multipath/NLoS scenarios, the proposed method is validated based on the positioning performance. Positioning solutions are computed by subtracting the estimated biases from the raw pseudoranges and comparing them to the ground truth.

A Scalable and Automated Framework for Tracking the Likely Adoption of Emerging Technologies

While new technologies are expected to revolutionise and become game-changers in improving the efficiency and practices of our daily lives, it is also critical to investigate and understand the barriers and opportunities faced by their adopters. Such findings can serve as an additional feature in the decisionmaking process when analysing the risks, costs, and benefits of adopting an emerging technology in a particular setting. Although several studies have attempted to perform such investigations, these approaches adopt a qualitative data collection methodology, which is limited in terms of the size of the targeted participant group and is associated with a significant manual overhead when transcribing and inferring results. This paper presents a scalable and automated framework for tracking the likely adoption and/or rejection of new technologies from a large landscape of adopters. In particular, a large corpus of social media texts containing references to emerging technologies was compiled. Text mining techniques were applied to extract the sentiments expressed towards technology aspects. In the context of the problem definition herein, we hypothesise that the expression of positive sentiment implies an increase in the likelihood of impacting a technology user’s acceptance to adopt, integrate, and/or use the technology, and negative sentiment implies an increase in the likelihood of impacting the rejection of emerging technologies by adopters. To quantitatively test our hypothesis, a ground truth analysis was performed to validate that the sentiments captured by the text mining approach were comparable to the results provided by human annotators when asked to label whether such texts positively or negatively impact their outlook towards adopting an emerging technology. The collected annotations demonstrated comparable results to those of the text mining approach, illustrating that the automatically extracted sentiments expressed towards technologies are useful features in understanding the landscape faced by technology adopters, as well as serving as an important decisionmaking component when, for example, recognising shifts in user behaviours, new demands, and emerging uncertainties.

Functional specialization of hippocampal somatostatin-expressing interneurons

Hippocampal somatostatin-expressing (Sst) GABAergic interneurons (INs) exhibit considerable anatomical and functional heterogeneity. Recent single-cell transcriptome analyses have provided a comprehensive Sst-IN subpopulations census, a plausible molecular ground truth of neuronal identity whose links to specific functionality remain incomplete. Here, we designed an approach to identify and access subpopulations of Sst-INs based on transcriptomic features. Four mouse models based on single or combinatorial Cre- and Flp- expression differentiated functionally distinct subpopulations of CA1 hippocampal Sst-INs that largely tiled the morpho-functional parameter space of the Sst-INs superfamily. Notably, the Sst;;Tac1 intersection revealed a population of bistratified INs that preferentially synapsed onto fast-spiking interneurons (FS-INs) and were sufficient to interrupt their firing. In contrast, the Ndnf;;Nkx2-1 intersection identified a population of oriens lacunosum-moleculare INs that predominantly targeted CA1 pyramidal neurons, avoiding FS-INs. Overall, our results provide a framework to translate neuronal transcriptomic identity into discrete functional subtypes that capture the diverse specializations of hippocampal Sst-INs.

Integrating meteorological and geospatial data for forest fire risk assessment

Understanding the factors that cause fire is crucial for minimizing the fire risk. In this research, a comprehensive approach was adopted to recognize factors influencing forest fires. Golestan National Park (GNP) was considered as a representative area with a humid climate in this study. Initially, using the Multi-Criteria Evaluation Method, a fire risk map was created by analyzing natural and human factors, and vulnerable areas were identified. Then, the relationship between key elements such as meteorological conditions, Land Surface Temperature (LST), and precipitation, with the occurrence of fire in different years was investigated. CHIRPS and Landsat data were utilized to assess LST changes and precipitation. 23-year changes in fire occurrence areas in GNP were acquired using MODIS products. The results of the data analysis showed that the highest number of fires occurred in forest areas, and in the fire risk prediction map, the extremely high-risk class is completely consistent with the ground truth data. The assigned weights, derived from expert opinions, highlight the substantial significance of elevation, and distance from roads and settlements. Additionally, the effectiveness of the model in providing reliable forecasts for fire risks in GNP is highlighted by the ROC curve with an AUC value of 0.83. Forest fires within GNP exhibit a distinct seasonality, predominantly occurring from July to December. During the warmer months, by coinciding with summer excursions, human activities may contribute to the ignition of fires. In 2013 and 2014, rising fire incidents correlated with elevated temperatures, hinting at a potential connection. GNP fires showed an upward trend with higher monthly LST and a downward trend with increased annual precipitation. The results showed that there is a relationship between LST, precipitation, and the occurrence of fire in GNP. Approximately 176.15 ha of GNP's forest areas have been destroyed by fires over the last two decades. This research demonstrated that there is a dynamic interaction between environmental conditions and fire incidents. By considering these factors, managers and environmental planners can develop effective strategies for managing and preventing forest fire risks.

Automated determination of the ion-recombination correction factor (ksat) in ultra-high dose rate electron radiation therapy.

Plane-parallel ionization chambers are the recommended secondary standard systems for clinical reference dosimetry of electrons. Dosimetry in high dose rate and dose-per-pulse (DPP) is challenging as ionization chambers are subject to ion recombination, especially when dose rate and/or DPP is increased beyond the range of conventional radiotherapy. The lack of universally accepted models for correction of ion recombination in UDHR is still an issue as it is, especially in FLASH-RT research, which is crucial in order to be able to accurately measure the dose for a wide range of dose rates and DPPs. The objective of this study was to show the feasibility of developing an Artificial Intelligence model to predict the ion-recombination factor-ksat for a plane-parallel Advanced Markus ionization chamber for conventional and ultra-high dose rate electron beams based on machine parameters. In addition, the predicted ksat of the AI model was compared with the current applied analytical models for this correction factor. A total number of 425 measurements was collected with a balanced variety in machine parameter settings. The specific ksat values were determined by dividing the output of the reference dosimeter (optically stimulated luminescence [OSL]) by the output of the AM chamber. Subsequently, a XGBoost regression model was trained, which used the different machine parameters as input features and the corresponding ksat value as output. The prediction accuracy of this regression model was characterized by R2-coefficient of determination, mean absolute error and root mean squared error. In addition, the model was compared with the Two-Voltage (TVA) method and empirical Petersson model for 19 different dose-per-pulse values ranging from conventional to UDHR regimes. The Akiake Information criterion (AIC) was calculated for the three different models. The XGBoost regression model reached a R2-score of 0.94 on the independent test set with a MAE of 0.067 and RMSE of 0.106. For the additional 19 random data points, the ksat values predicted by the XGBoost model showed to be in agreement, within the uncertainties, with the ones determined by the Petersson model and better than the TVA method for doses per pulse>3.5Gy with a maximum deviation from the ground truth of 14.2%, 16.7%, and -36.0%, respectively, for DPP>4Gy. The proposed method of using AI for ksat determination displays efficiency. For the investigated DPPs, the ksat values obtained with the XGBoost model were in concurrence with the ones obtained with the current available analytical models within the boundaries of uncertainty, certainly for the DPP characterizing UDHR. But the overall performance of the AI model, taking the number of free parameters into account, lacked efficiency. Future research should optimize the determination of the experimental ksat, and investigate the determination the ksat for DPPs higher than the ones investigated in this study, while also evaluating the prediction of the proposed XGBoost model for UDHR machines of different centers.

Deep learning-based optimization of field geometry for total marrow irradiation delivered with volumetric modulated arc therapy.

Total marrow (lymphoid) irradiation (TMI/TMLI) is a radiotherapy treatment used to selectively target the bone marrow and lymph nodes in conditioning regimens for allogeneic hematopoietic stem cell transplantation. A complex field geometry is needed to cover the large planning target volume (PTV) of TMI/TMLI with volumetric modulated arc therapy (VMAT). Five isocenters and ten overlapping fields are needed for the upper body, while, for patients with large anatomical conformation, two specific isocenters are placed on the arms. The creation of a field geometry is clinically challenging and is performed by a medical physicist (MP) specialized in TMI/TMLI. To develop convolutional neural networks (CNNs) for automatically generating the field geometry of TMI/TMLI. The dataset comprised 117 patients treated with TMI/TMLI between 2011 and 2023 at our Institute. The CNN input image consisted of three channels, obtained by projecting along the sagittal plane: (1) average CT pixel intensity within the PTV; (2) PTV mask; (3) brain, lungs, liver, bowel, and bladder masks. This "averaged" frontal view combined the information analyzed by the MP when setting the field geometry in the treatment planning system (TPS). Two CNNs were trained to predict the isocenters coordinates and jaws apertures for patients with (CNN-1) and without (CNN-2) isocenters on the arms. Local optimization methods were used to refine the models output based on the anatomy of the patient. Model evaluation was performed on a test set of 15 patients in two ways: (1) by computing the root mean squared error (RMSE) between the CNN output and ground truth; (2) with a qualitative assessment of manual and generated field geometries-scale: 1=not adequate, 4=adequate-carried out in blind mode by three MPs with different expertise in TMI/TMLI. The Wilcoxon signed-rank test was used to evaluate the independence of the given scores between manual and generated configurations (p<0.05 significant). The average and standard deviation values of RMSE for CNN-1 and CNN-2 before/after local optimization were 15±2/13±3mm and 16±2/18±4mm, respectively. The CNNs were integrated into a planning automation software for TMI/TMLI such that the MPs could analyze in detail the proposed field geometries directly in the TPS. The selection of the CNN model to create the field geometry was based on the PTV width to approximate the decision process of an experienced MP and provide a single option of field configuration. We found no significant differences between the manual and generated field geometries for any MP, with median values of 4versus 4 (p=0.92), 3versus 3 (p=0.78), 4versus 3 (p=0.48), respectively. Starting from October 2023, the generated field geometry has been introduced in our clinical practice for prospective patients. The generated field geometries were clinically acceptable and adequate, even for an MP with high level of expertise in TMI/TMLI. Incorporating the knowledge of the MPs into the development cycle was crucial for optimizing the models, especially in this scenario with limited data.

Physics-informed neural networks for acoustic boundary admittance estimation

Acoustic simulations often face significant uncertainties due to limited knowledge of acoustic boundary conditions. While measuring the boundary admittance in situ is challenging in practical applications, numerical inverse methods can be used to characterize the boundary conditions based on sound pressure data. However, conventional inverse methods require a validated forward model and can become impractical for computationally expensive simulation models. Over the past years, machine learning approaches have emerged as promising methods for scientific computing and data-driven modeling. Physics-informed neural networks incorporate physical prior knowledge into a neural network by adding the residual of the underlying partial differential equation to the loss function. Training the neural network minimizes the loss function, allowing the network to learn a solution that not only fits the training data but also satisfies the corresponding boundary value problem. In this study, physics-informed neural networks are trained to learn the sound pressure solution within two numerical examples governed by the Helmholtz equation without explicitly specifying the boundary conditions at selected boundaries. After training, the neural network’s prediction of the boundary admittance is evaluated and compared to the ground truth, initially assigned in the finite element reference solution. Additionally, the proposed method is validated using experimental data obtained from an acoustic impedance tube measurement. The results show that physics-informed neural networks can accurately learn the sound pressure field and implicitly solve the inverse problem by providing an accurate estimate of the underlying boundary admittance, even in the case of spatially varying boundary conditions.

Identifiability of spatiotemporal tissue perfusion models.

Standard models for perfusion quantification in DCE-MRI produce a bias by treating voxels as isolated systems. Spatiotemporal models can remove this bias, but it is unknown whether they are fundamentally identifiable. The aim of this study is to investigate this question in-silico using one-dimensional toy systems with a one-compartment blood flow model and a two-compartment perfusion model.&#xD;&#xD;Approach: For each of the two models, identifiability is explored theoretically and in-silico for three systems. Concentrations over space and time are simulated by forward propagation. Different levels of noise and temporal undersampling are added to investigate sensitivity to measurement error. Model parameters are fitted using a standard gradient descent algorithm, applied iteratively with a stepwise increasing time window. Model fitting is repeated with different initial values to probe uniqueness of the solution. Reconstruction accuracy is quantified for each parameter by comparison to the ground truth. &#xD;&#xD;Main Results: Theoretical analysis shows that flows and volume fractions are only identifiable up to a constant, and that this degeneracy can be removed by proper choice of parameters. Simulations show that in all cases, the tissue concentrations can be reconstructed accurately. The one-compartment model shows accurate reconstruction of blood velocities and arterial input functions, independent of the initial values and robust to measurement error. The two-compartmental perfusion model was not fully identifiable, showing good reconstruction of arterial velocities and input functions, but multiple valid solutions for the perfusion parameters and venous velocities, and a strong sensitivity to measurement error in these parameters. &#xD;&#xD;Significance:These results support the use of one-compartment spatiotemporal flow models, but two-compartment perfusion models were not sufficiently identifiable. Future studies should investigate whether this degeneracy is resolved in more realistic 2D and 3D systems, by adding physically justified constraints, or by optimizing experimental parameters such as injection duration or temporal resolution.

Three-dimensional biphase fabric estimation from 2D images by deep learning

A pruned VGG19 model subjected to Axial Coronal Sagittal (ACS) convolutions and a custom VGG16 model are benchmarked to predict 3D fabric descriptors from a set of 2D images. The data used for training and testing are extracted from a set of 600 3D biphase microstructures created numerically. Fabric descriptors calculated from the 3D microstructures constitute the ground truth, while the input data are obtained by slicing the 3D microstructures in each direction of space at regular intervals. The computational cost to train the custom ACS-VGG19 model increases linearly with p (the number of images extracted in each direction of space), and increasing p does not improve the performance of the model - or only does so marginally. The best performing ACS-VGG19 model provides a MAPE of 2 to 5% for the means of aggregate size, aspect ratios and solidity, but cannot be used to estimate orientations. The custom VGG16 yields a MAPE of 2% or less for the means of aggregate size, distance to nearest neighbor, aspect ratios and solidity. The MAPE is less than 3% for the mean roundness, and in the range of 5-7% for the aggregate volume fraction and the mean diagonal components of the orientation matrix. Increasing p improves the performance of the custom VGG16 model, but becomes cost ineffective beyond 3 images per direction. For both models, the aggregate volume fraction is predicted with less accuracy than higher order descriptors, which is attributed to the bias given by the loss function towards highly-correlated descriptors. Both models perform better to predict means than standard deviations, which are noisy quantities. The custom VGG16 model performs better than the pruned version of the ACS-VGG19 model, likely because it contains 3 times (p = 1) to 28 times (p = 10) less parameters than the ACS-VGG19 model, allowing better and faster cnvergence, with less data. The custom VGG16 model predicts the second and third invariants of the orientation matrix with a MAPE of 2.8% and 8.9%, respectively, which suggests that the model can predict orientation descriptors regardless of the orientation of the input images.

A deep learning-based workflow for fast prediction of 3D state variables in geological carbon storage: A dimension reduction approach

Deep learning (DL) models are extensively used as surrogate models for high-fidelity simulations of multiphase fluid flow in porous media at large scales, enabling fast forecasts of the spatial–temporal evolution of three-dimensional (3D) state variables in geological carbon storage (GCS). However, training these models in high-dimensional space remains computationally demanding and prone to overfitting because of limited training data. This paper presents a novel workflow to address these challenges by integrating dimension reduction (DR) methods. The proposed workflow employed pre-trained DR models to extract the latent variables of geological models and state variables and utilized the multi-layer perceptron (MLP) for constructing mapping functions between the input and output variables in latent spaces. Subsequently, the pre-trained reconstruction models converted the MLP-predicted latent state variables to their original high-dimensional form. Furthermore, we proposed a novel strategy for the DR and reconstruction of 3D saturation fields to account for the unique data characteristics of sparsity, nonuniformity, and discontinuity. The proposed strategy applied PCA and inverse PCA for 2D average saturation fields and developed a DL-based 3D reconstruction model, leveraging three 2D average saturation fields as input to produce a 3D saturation field as output. The pre-training of DR and reconstruction models and training of MLP models were conducted on 84 Gulf of Mexico (GoM) simulations and evaluated on 12 testing simulations. Each simulation contained 720 monthly time steps, with the first 360 months as the injection period and the rest as the post-injection period. The proposed workflow, incorporating DR and DL models, accurately predicts the normalized 3D pressure fields, achieving mean square error (MSE) of 2.92 × 10−7 compared to the ground truth obtained from a full-physics simulator. Furthermore, the proposed strategy outperformed PCA and convolutional autoencoder (CAE) models on 3D saturation fields, resulting in minor workflow prediction errors with an MSE of 2.93 × 10−5. The results suggest the proposed workflow provides sufficient predictive fidelity across temporal and spatial scales, and enables a speedup of 160 times compared to the full-physics simulator, facilitating improved decision-making and risk assessment for large-scale GCS management in real-time scenarios.

A Machine Learning System to Indicate Diagnosis of Idiopathic Pulmonary Fibrosis Non-Invasively in Challenging Cases.

Radiologic usual interstitial pneumonia (UIP) patterns and concordant clinical characteristics define a diagnosis of idiopathic pulmonary fibrosis (IPF). However, limited expert access and high inter-clinician variability challenge early and pre-invasive diagnostic sensitivity and differentiation of IPF from other interstitial lung diseases (ILDs). We investigated a machine learning-driven software system, Fibresolve, to indicate IPF diagnosis in a heterogeneous group of 300 patients with interstitial lung disease work-up in a retrospective analysis of previously and prospectively collected registry data from two US clinical sites. Fibresolve analyzed cases at the initial pre-invasive assessment. An Expert Clinical Panel (ECP) and three panels of clinicians with varying experience analyzed the cases for comparison. Ground Truth was defined by separate multi-disciplinary discussion (MDD) with the benefit of surgical pathology results and follow-up. Fibresolve met both pre-specified co-primary endpoints of sensitivity superior to ECP and significantly greater specificity (p = 0.0007) than the non-inferior boundary of 80.0%. In the key subgroup of cases with thin-slice CT and atypical UIP patterns (n = 124), Fibresolve's diagnostic yield was 53.1% [CI: 41.3-64.9] (versus 0% pre-invasive clinician diagnostic yield in this group), and its specificity was 85.9% [CI: 76.7-92.6%]. Overall, Fibresolve was found to increase the sensitivity and diagnostic yield for IPF among cases of patients undergoing ILD work-up. These results demonstrate that in combination with standard clinical assessment, Fibresolve may serve as an adjunct in the diagnosis of IPF in a pre-invasive setting.

Weakly Supervised Exaggeration Transfer for Caricature Generation With Cross-Modal Knowledge Distillation.

Caricature generation aims to translate portrait photos into caricatures with exaggerated and hand-drawn artistic styles. Previous methods faced challenges in creating diverse and meaningful exaggeration effects, yielding unsatisfactory and uncontrollable results. To overcome this, we proposed ETCari, a novel weakly supervised exaggeration transfer network. ETCari enables the learning of diverse exaggeration caricature styles from various artists, better meeting individual customization requirements and achieving diversified exaggeration while retaining identity features. Specifically, we use the thin-plate spline control point deformation field as the ground truth, serving as the loss for weakly supervised learning to address the challenge of no labels. We convert input to an intermediate modality for domain adaptation, training a teacher model. Subsequently, we perform cross-modal knowledge distillation to train the student model, simplifying preprocessing during inference and avoiding the impact of face parser errors. Experiments on the WebCaricature dataset demonstrate that ETCari effectively performs exaggeration transfer, generating appealing caricatures.

A contrastive learning approach to integrate spatial transcriptomics and histological images

The rapid growth of spatially resolved transcriptomics technology provides new perspectives on spatial tissue architecture. Deep learning has been widely applied to derive useful representations for spatial transcriptome analysis. However, effectively integrating spatial multi-modal data remains challenging. Here, we present ConGcR, a contrastive learning-based model for integrating gene expression, spatial location, and tissue morphology for data representation and spatial tissue architecture identification. Graph convolution and ResNet were used as encoders for gene expression with spatial location and histological image inputs, respectively. We further enhanced ConGcR with a graph auto-encoder as ConGaR to better model spatially embedded representations. We validated our models using 16 human brains, four chicken hearts, eight breast tumors, and 30 human lung spatial transcriptomics samples. The results showed that our models generated more effective embeddings for obtaining tissue architectures closer to the ground truth than other methods. Overall, our models not only can improve tissue architecture identification’s accuracy but also may provide valuable insights and effective data representation for other tasks in spatial transcriptome analyses.

Predicting treatment plan approval probability for high-dose-rate brachytherapy of cervical cancer using adversarial deep learning

Objective. Predicting the probability of having the plan approved by the physician is important for automatic treatment planning. Driven by the mathematical foundation of deep learning that can use a deep neural network to represent functions accurately and flexibly, we developed a deep-learning framework that learns the probability of plan approval for cervical cancer high-dose-rate brachytherapy (HDRBT). Approach. The system consisted of a dose prediction network (DPN) and a plan-approval probability network (PPN). DPN predicts organs at risk (OAR) D 2cc and CTV D 90% of the current fraction from the patient’s current anatomy and prescription dose of HDRBT. PPN outputs the probability of a given plan being acceptable to the physician based on the patients anatomy and the total dose combining HDRBT and external beam radiotherapy sessions. Training of the networks was achieved by first training them separately for a good initialization, and then jointly via an adversarial process. We collected approved treatment plans of 248 treatment fractions from 63 patients. Among them, 216 plans from 54 patients were employed in a four-fold cross validation study, and the remaining 32 plans from other 9 patients were saved for independent testing. Main results. DPN predicted equivalent dose of 2 Gy for bladder, rectum, sigmoid D 2cc and CTV D 90% with a relative error of 11.51% ± 6.92%, 8.23% ± 5.75%, 7.12% ± 6.00%, and 10.16% ± 10.42%, respectively. In a task that differentiates clinically approved plans and disapproved plans generated by perturbing doses in ground truth approved plans by 20%, PPN achieved accuracy, sensitivity, specificity, and area under the curve 0.70, 0.74, 0.65, and 0.74. Significance. We demonstrated the feasibility of developing a novel deep-learning framework that predicts a probability of plan approval for HDRBT of cervical cancer, which is an essential component in automatic treatment planning.

High-resolution 3T to 7T ADC map synthesis with a hybrid CNN-transformer model.

7 Tesla (7T) apparent diffusion coefficient (ADC) maps derived from diffusion-weighted imaging (DWI) demonstrate improved image quality and spatial resolution over 3 Tesla (3T) ADC maps. However, 7T magnetic resonance imaging (MRI) currently suffers from limited clinical unavailability, higher cost, and increased susceptibility to artifacts. To address these issues, we propose a hybrid CNN-transformer model to synthesize high-resolution 7T ADC maps from multimodal 3T MRI. The Vision CNN-Transformer (VCT), composed of both Vision Transformer (ViT) blocks and convolutional layers, is proposed to produce high-resolution synthetic 7T ADC maps from 3T ADC maps and 3T T1-weighted (T1w) MRI. ViT blocks enabled global image context while convolutional layers efficiently captured fine detail. The VCT model was validated on the publicly available Human Connectome Project Young Adult dataset, comprising 3T T1w, 3T DWI, and 7T DWI brain scans. The Diffusion Imaging in Python library was used to compute ADC maps from the DWI scans. A total of 171 patient cases were randomly divided into 130 training cases, 20 validation cases, and 21 test cases. The synthetic ADC maps were evaluated by comparing their similarity to the ground truth volumes with the following metrics: peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and mean squared error (MSE). In addition, RESULTS: The results are as follows: PSNR: 27.0±0.9dB, SSIM: 0.945±0.010, and MSE: 2.0E-3±0.4E-3. Both qualitative and quantitative results demonstrate that VCT performs favorably against other state-of-the-art methods. We have introduced various efficiency improvements, including the implementation of flash attention and training on 176×208 resolution images. These enhancements have resulted in the reduction of parameters and training time per epoch by 50% in comparison to ResViT. Specifically, the training time per epoch has been shortened from 7.67 min to 3.86 min. We propose a novel method to predict high-resolution 7T ADC maps from low-resolution 3T ADC maps and T1w MRI. Our predicted images demonstrate better spatial resolution and contrast compared to 3T MRI and prediction results made by ResViT and pix2pix. These high-quality synthetic 7T MR images could be beneficial for disease diagnosis and intervention, producing higher resolution and conformal contours, and as an intermediate step in generating synthetic CT for radiation therapy, especially when 7T MRI scanners are unavailable.

A Geospatial Analysis of Environmental Impacts of Climate Change on Osogbo and Environs

Climate change is one of the alarming global environmental changes likely to have deleterious effects on natural, social, cultural, and human systems. The risks associated with it call for a broad spectrum of policy responses and strategies at local, regional, national, and global levels. This research aims to carry out a geospatial analysis of the environmental impact of climate change on Osogbo and its environs with a view to providing information towards mitigating the negative impact of climate change in the study area. This study uses Landsat 4, Landsat 7, and Landsat 8 satellite data from 1982, 2002, and 2022, respectively, to evaluate land cover and temperature variation as a climatic change in the study area over four decades. The land survey method was carried out to collect ground truth data for validation of the satellite imageries. The drought map was produced based on the Vegetation Condition Index (VCI) map computed from Normalized Difference Vegetation Index (NDVI) datasets and was further classified into three categories: moderate, mild, and no drought. Analysis of the Land Use Land Cover (LULC) of the study area shows that the vegetation decreased from 84% in 1982 to 59% in 2002 and 56% in 2022. The built-up/exposed surface increased from 15.28% in 1982 to 38% in 2002 and 41.28% in 2022. The water bodies also increased from 0.17% in 1982 to 15.31% in 2022. LST analysis reveals that the mean temperature of the study area increased by 10.73 °C during the past four decades (from 1982 to 2022), with the most accelerated warming (11.41 °C) occurring during the last two decades (from 2002 to 2022). The Vegetation Condition Index (VCI) based drought map shows that there was no extreme or severe drought in the study area. However, moderate and mild droughts were observed mostly in the developed part of the study area, which confirmed the presence of drought. It is hereby recommended to introduce sustainable water management strategies to mitigate water shortage in the study area.

Look4LTRs: a Long terminal repeat retrotransposon detection tool capable of cross species studies and discovering recently nested repeats

Plant genomes include large numbers of transposable elements. One particular type of these elements is flanked by two Long Terminal Repeats (LTRs) and can translocate using RNA. Such elements are known as LTR-retrotransposons; they are the most abundant type of transposons in plant genomes. They have many important functions involving gene regulation and the rise of new genes and pseudo genes in response to severe stress. Additionally, LTR-retrotransposons have several applications in biotechnology. Due to the abundance and the importance of LTR-retrotransposons, multiple computational tools have been developed for their detection. However, none of these tools take advantages of the availability of related genomes; they process one chromosome at a time. Further, recently nested LTR-retrotransposons (multiple elements of the same family are inserted into each other) cannot be annotated accurately — or cannot be annotated at all — by the currently available tools. Motivated to overcome these two limitations, we built Look4LTRs, which can annotate LTR-retrotransposons in multiple related genomes simultaneously and discover recently nested elements. The methodology of Look4LTRs depends on techniques imported from the signal-processing field, graph algorithms, and machine learning with a minimal use of alignment algorithms. Four plant genomes were used in developing Look4LTRs and eight plant genomes for evaluating it in contrast to three related tools. Look4LTRs is the fastest while maintaining better or comparable F1 scores (the harmonic average of recall and precision) to those obtained by the other tools. Our results demonstrate the added benefit of annotating LTR-retrotransposons in multiple related genomes simultaneously and the ability to discover recently nested elements. Expert human manual examination of six elements — not included in the ground truth — revealed that three elements belong to known families and two elements are likely from new families. With respect to examining recently nested LTR-retrotransposons, three out of five were confirmed to be valid elements. Look4LTRs — with its speed, accuracy, and novel features — represents a true advancement in the annotation of LTR-retrotransposons, opening the door to many studies focused on understanding their functions in plants.