MAGMA. 2024 Aug 8. doi: 10.1007/s10334-024-01189-0. Online ahead of print.

ABSTRACT

INTRODUCTION: Quantification of dynamic contrast-enhanced (DCE)-MRI has the potential to provide valuable clinical information, but robust pharmacokinetic modeling remains a challenge for clinical adoption.

METHODS: A 7-layer neural network called DCE-Qnet was trained on simulated DCE-MRI signals derived from the Extended Tofts model with the Parker arterial input function. Network training incorporated B1 inhomogeneities to estimate perfusion (Ktrans, vp, ve), tissue T1 relaxation, proton density and bolus arrival time (BAT). The accuracy was tested in a digital phantom in comparison to a conventional nonlinear least-squares fitting (NLSQ). In vivo testing was conducted in ten healthy subjects. Regions of interest in the cervix and uterine myometrium were used to calculate the inter-subject variability. The clinical utility was demonstrated on a cervical cancer patient. Test-retest experiments were used to assess reproducibility of the parameter maps in the tumor.

RESULTS: The DCE-Qnet reconstruction outperformed NLSQ in the phantom. The coefficient of variation (CV) in the healthy cervix varied between 5 and 51% depending on the parameter. Parameter values in the tumor agreed with previous studies despite differences in methodology. The CV in the tumor varied between 1 and 47%.

CONCLUSION: The proposed approach provides comprehensive DCE-MRI quantification from a single acquisition. DCE-Qnet eliminates the need for separate T1 scan or BAT processing, leading to a reduction of 10 min per scan and more accurate quantification.

PMID:39112813 | DOI:10.1007/s10334-024-01189-0

J Plant Res. 2024 Aug 7. doi: 10.1007/s10265-024-01565-w. Online ahead of print.

ABSTRACT

The vascular bundle in the ear-internode of maize is a key conduit for transporting photosynthetic materials between "source" and "sink", making it critically important to examine its micro-phenotypes and genetic architecture to identify advantageous characteristics and cultivate high-yielding and high-quality varieties. Unfortunately, the limited observation methods and scope of study precludes any comprehensive and systematic investigations into the microscopic phenotypes and genetic mechanisms of vascular bundle in maize ear-internode. In this study, 47 phenotypic traits were extracted in 495 maize inbred lines using micro computed tomography (Micro-CT) scanning technology and a deep learning-based phenotype acquisition method for stem vascular bundle, which included stem slice-related, epidermis zone-related, periphery zone-related, inner zone-related and vascular bundles-related traits. Phenotypic analysis indicated that there was extensive phenotypic variation of vascular bundle traits in ear-internode, especially that in the inner zone. Of these, 30 phenotypic traits with heritability greater than 0.70 were conducted for GWAS, and a total of 4,225 significant SNPs and 416 candidate genes with detailed functional annotations were identified. Furthermore, 20 genes were highly expressed in stem-related tissues, especially in maize internodes. Functional analysis of candidate genes indicated that the pathways obtained for candidate genes of different trait groups were distinct, mainly involved in vitamin synthesis and metabolism, transport of substances, carbohydrate derivative catabolic process, protein transport and localization, and anatomical structure development. The results of this study will help to further understand the phenotypic traits of stem vascular bundles and provide a reference for revealing the genetic mechanism of maize ear-internode vascular bundles.

PMID:39112806 | DOI:10.1007/s10265-024-01565-w

Commun Biol. 2024 Aug 7;7(1):954. doi: 10.1038/s42003-024-06648-x.

ABSTRACT

Parkinson's disease (PD) exhibits heterogeneity in terms of symptoms and prognosis, likely due to diverse neuroanatomical alterations. This study employs a contrastive deep learning approach to analyze Magnetic Resonance Imaging (MRI) data from 932 PD patients and 366 controls, aiming to disentangle PD-specific neuroanatomical alterations. The results reveal that these neuroanatomical alterations in PD are correlated with individual differences in dopamine transporter binding deficit, neurodegeneration biomarkers, and clinical severity and progression. The correlation with clinical severity is verified in an external cohort. Notably, certain proteins in the cerebrospinal fluid are strongly associated with PD-specific features, particularly those involved in the immune function. The most notable neuroanatomical alterations are observed in both subcortical and temporal regions. Our findings provide deeper insights into the patterns of brain atrophy in PD and potential underlying molecular mechanisms, paving the way for earlier patient stratification and the development of treatments to slow down neurodegeneration.

PMID:39112797 | DOI:10.1038/s42003-024-06648-x

Sci Rep. 2024 Aug 7;14(1):18329. doi: 10.1038/s41598-024-69405-4.

ABSTRACT

We developed a surgical support system that visualises important microanatomies using artificial intelligence (AI). This study evaluated its accuracy in recognising the thoracic nerves during lung cancer surgery. Recognition models were created with deep learning using images precisely annotated for nerves. Computational evaluation was performed using the Dice index and the Jaccard index. Four general thoracic surgeons evaluated the accuracy of nerve recognition. Further, the differences in time lag, image quality and smoothness of movement between the AI system and surgical monitor were assessed. Ratings were made using a five-point scale. The computational evaluation was relatively favourable, with a Dice index of 0.56 and a Jaccard index of 0.39. The AI system was used for 10 thoracoscopic surgeries for lung cancer. The accuracy of thoracic nerve recognition was satisfactory, with a recall score of 4.5 ± 0.4 and a precision score of 4.0 ± 0.9. Though smoothness of motion (3.2 ± 0.4) differed slightly, nearly no difference in time lag (4.9 ± 0.3) and image quality (4.6 ± 0.5) between the AI system and the surgical monitor were observed. In conclusion, the AI surgical support system has a satisfactory accuracy in recognising the thoracic nerves.

PMID:39112794 | DOI:10.1038/s41598-024-69405-4

Skeletal Radiol. 2024 Aug 8. doi: 10.1007/s00256-024-04769-2. Online ahead of print.

ABSTRACT

OBJECTIVE: To evaluate the diagnostic performance and image quality of accelerated Turbo Spin Echo sequences using deep-learning (DL) reconstructions compared to conventional sequences in knee and ankle MRIs of children and young adults.

MATERIALS AND METHODS: IRB-approved prospective study consisting of 49 MRIs from 48 subjects (10 males, mean age 16.4 years, range 7-29 years), with each MRI consisting of both conventional and DL sequences. Sequences were evaluated blindly to determine predictive values, sensitivity, and specificity of DL sequences using conventional sequences and knee arthroscopy (if available) as references. Physeal patency and appearance were evaluated. Qualitative parameters were compared. Presence of undesired image alterations was assessed.

RESULTS: The prevalence of abnormal findings in the knees and ankles were 11.7% (75/640), and 11.5% (19/165), respectively. Using conventional sequences as reference, sensitivity and specificity of DL sequences in knees were 90.7% and 99.3%, and in ankles were 100.0% and 100.0%. Using arthroscopy as reference, sensitivity and specificity of DL sequences were 80.0% and 95.8%, and of conventional sequences were 80.0% and 97.9%. Agreement of physeal status was 100.0%. DL sequences were qualitatively "same-or-better" compared to conventional (p < 0.032), except for pixelation artifact for the PDFS sequence (p = 0.233). No discrete image alteration was identified in the knee DL sequences. In the ankle, we identified one DL artifact involving a tendon (0.8%, 1/125). DL sequences were faster than conventional sequences by a factor of 2 (p < 0.001).

CONCLUSION: In knee and ankle MRIs, DL sequences provided similar diagnostic performance and "same-or-better" image quality than conventional sequences at half the acquisition time.

PMID:39112675 | DOI:10.1007/s00256-024-04769-2

J Environ Manage. 2024 Aug 6;367:122018. doi: 10.1016/j.jenvman.2024.122018. Online ahead of print.

ABSTRACT

Climate change has a significant impact on dissolved oxygen (DO) concentrations, particularly in coastal inlets where numerous human activities occur. Due to the various water quality (WQ), hydrological, and climatic parameters that influence this phenomenon, predicting and modeling DO variation is a challenging process. Accordingly, this study introduces an innovative Deep Learning Neural Network (DLNN) methodology to model and predict DO concentrations for the Egyptian Rashid coastal inlet, leveraging field-recorded WQ and hydroclimatic datasets. Initially, statistical and exploratory data analyses are performed to provide a thorough understanding of the relationship between DO fluctuations and associated WQ and hydroclimatic stressors. As an initial step towards developing an effective DO predictive model, conventional Machine Learning (ML) approaches such as Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Decision Tree Regressor (DTR) are employed. Subsequently, a DLNN approach is utilized to validate the prediction capabilities of the investigated conventional ML approaches. Finally, a sensitivity analysis is conducted to evaluate the impact of WQ and hydroclimatic parameters on predicted DO. The outcomes demonstrate that DLNN significantly improves DO prediction accuracy by 4% compared to the best-performing ML approach, achieving a Correlation Coefficient of 0.95 with a root mean square error (RMSE) of 0.42 mg/l. Solar radiation (SR), pH, water levels (WL), and atmospheric pressure (P) emerge as the most significant hydroclimatic parameters influencing DO fluctuations. Ultimately, the developed models could serve as effective indicators for coastal authorities to monitor DO changes resulting from accelerated climate change along the Egyptian coast.

PMID:39111007 | DOI:10.1016/j.jenvman.2024.122018

PLoS Comput Biol. 2024 Aug 7;20(8):e1012329. doi: 10.1371/journal.pcbi.1012329. Online ahead of print.

ABSTRACT

Our understanding of bird song, a model system for animal communication and the neurobiology of learning, depends critically on making reliable, validated comparisons between the complex multidimensional syllables that are used in songs. However, most assessments of song similarity are based on human inspection of spectrograms, or computational methods developed from human intuitions. Using a novel automated operant conditioning system, we collected a large corpus of zebra finches' (Taeniopygia guttata) decisions about song syllable similarity. We use this dataset to compare and externally validate similarity algorithms in widely-used publicly available software (Raven, Sound Analysis Pro, Luscinia). Although these methods all perform better than chance, they do not closely emulate the avian assessments. We then introduce a novel deep learning method that can produce perceptual similarity judgements trained on such avian decisions. We find that this new method outperforms the established methods in accuracy and more closely approaches the avian assessments. Inconsistent (hence ambiguous) decisions are a common occurrence in animal behavioural data; we show that a modification of the deep learning training that accommodates these leads to the strongest performance. We argue this approach is the best way to validate methods to compare song similarity, that our dataset can be used to validate novel methods, and that the general approach can easily be extended to other species.

PMID:39110762 | DOI:10.1371/journal.pcbi.1012329

Cell Rep. 2024 Aug 6;43(8):114583. doi: 10.1016/j.celrep.2024.114583. Online ahead of print.

ABSTRACT

Vast shotgun metagenomics data remain an underutilized resource for novel enzymes. Artificial intelligence (AI) has increasingly been applied to protein mining, but its conventional performance evaluation is interpolative in nature, and these trained models often struggle to extrapolate effectively when challenged with unknown data. In this study, we present a framework (DeepMineLys [deep mining of phage lysins from human microbiome]) based on the convolutional neural network (CNN) to identify phage lysins from three human microbiome datasets. When validated with an independent dataset, our method achieved an F1-score of 84.00%, surpassing existing methods by 20.84%. We expressed 16 lysin candidates from the top 100 sequences in E. coli, confirming 11 as active. The best one displayed an activity 6.2-fold that of lysozyme derived from hen egg white, establishing it as the most potent lysin from the human microbiome. Our study also underscores several important issues when applying AI to biology questions. This framework should be applicable for mining other proteins.

PMID:39110597 | DOI:10.1016/j.celrep.2024.114583

Transl Vis Sci Technol. 2024 Aug 1;13(8):11. doi: 10.1167/tvst.13.8.11.

ABSTRACT

PURPOSE: To predict 10-2 Humphrey visual fields (VFs) from 24-2 VFs and associated non-total deviation features using deep learning.

METHODS: We included 5189 reliable 24-2 and 10-2 VF pairs from 2236 patients, and 28,409 reliable pairs of macular OCT scans and 24-2 VF from 19,527 eyes of 11,560 patients. We developed a transformer-based deep learning model using 52 total deviation values and nine VF test features to predict 68 10-2 total deviation values. The mean absolute error, root mean square error, and the R2 were evaluation metrics. We further evaluated whether the predicted 10-2 VFs can improve the structure-function relationship between macular thinning and paracentral VF loss in glaucoma.

RESULTS: The average mean absolute error and R2 for 68 10-2 VF test points were 3.30 ± 0.52 dB and 0.70 ± 0.11, respectively. The accuracy was lower in the inferior temporal region. The model placed greater emphasis on 24-2 VF points near the central fixation point when predicting the 10-2 VFs. The inclusion of nine VF test features improved the mean absolute error and R2 up to 0.17 ± 0.06 dB and 0.01 ± 0.01, respectively. Age was the most important 24-2 VF test parameter for 10-2 VF prediction. The predicted 10-2 VFs achieved an improved structure-function relationship between macular thinning and paracentral VF loss, with the R2 at the central 4, 12, and 16 locations of 24-2 VFs increased by 0.04, 0.05 and 0.05, respectively (P < 0.001).

CONCLUSIONS: The 10-2 VFs may be predicted from 24-2 data.

TRANSLATIONAL RELEVANCE: The predicted 10-2 VF has the potential to improve glaucoma diagnosis.

PMID:39110574 | DOI:10.1167/tvst.13.8.11

IEEE Trans Image Process. 2024 Aug 7;PP. doi: 10.1109/TIP.2024.3437219. Online ahead of print.

ABSTRACT

A common approach to solve inverse imaging problems relies on finding a maximum a posteriori (MAP) estimate of the original unknown image, by solving a minimization problem. In this context, iterative proximal algorithms are widely used, enabling to handle non-smooth functions and linear operators. Recently, these algorithms have been paired with deep learning strategies, to further improve the estimate quality. In particular, proximal neural networks (PNNs) have been introduced, obtained by unrolling a proximal algorithm as for finding a MAP estimate, but over a fixed number of iterations, with learned linear operators and parameters. As PNNs are based on optimization theory, they are very flexible, and can be adapted to any image restoration task, as soon as a proximal algorithm can solve it. They further have much lighter architectures than traditional networks. In this article we propose a unified framework to build PNNs for the Gaussian denoising task, based on both the dual-FB and the primal-dual Chambolle-Pock algorithms. We further show that accelerated inertial versions of these algorithms enable skip connections in the associated NN layers. We propose different learning strategies for our PNN framework, and investigate their robustness (Lipschitz property) and denoising efficiency. Finally, we assess the robustness of our PNNs when plugged in a forward-backward algorithm for an image deblurring problem.

PMID:39110565 | DOI:10.1109/TIP.2024.3437219

IEEE J Biomed Health Inform. 2024 Aug 7;PP. doi: 10.1109/JBHI.2024.3440171. Online ahead of print.

ABSTRACT

Objective - Medical image segmentation is essential for several clinical tasks, including diagnosis, surgical and treatment planning, and image-guided interventions. Deep Learning (DL) methods have become the state-of-the-art for several image segmentation scenarios. However, a large and well-annotated dataset is required to effectively train a DL model, which is usually difficult to obtain in clinical practice, especially for 3D images. Methods - In this paper, we proposed Deep-DM, a learning-guided deformable model framework for 3D medical imaging segmentation using limited training data. In the proposed method, an energy function is learned by a Convolutional Neural Network (CNN) and integrated into an explicit deformable model to drive the evolution of an initial surface towards the object to segment. Specifically, the learning-based energy function is iteratively retrieved from localized anatomical representations of the image containing the image information around the evolving surface at each iteration. By focusing on localized regions of interest, this representation excludes irrelevant image information, facilitating the learning process. Results and conclusion - The performance of the proposed method is demonstrated for the tasks of left ventricle and fetal head segmentation in ultrasound, left atrium segmentation in Magnetic Resonance, and bladder segmentation in Computed Tomography, using different numbers of training volumes in each study. The results obtained showed the feasibility of the proposed method to segment different anatomical structures in different imaging modalities. Moreover, the results also showed that the proposed approach is less dependent on the size of the training dataset in comparison with state-of-the-art DL-based segmentation methods, outperforming them for all tasks when a low number of samples is available. Significance - Overall, by offering a more robust and less data-intensive approach to accurately segmenting anatomical structures, the proposed method has the potential to enhance clinical tasks that require image segmentation strategies.

PMID:39110559 | DOI:10.1109/JBHI.2024.3440171

Cardiovasc Intervent Radiol. 2024 Aug 7. doi: 10.1007/s00270-024-03830-w. Online ahead of print.

NO ABSTRACT

PMID:39110204 | DOI:10.1007/s00270-024-03830-w

J Chem Theory Comput. 2024 Aug 7. doi: 10.1021/acs.jctc.4c00385. Online ahead of print.

ABSTRACT

With the recent introduction of deep learning techniques into the prediction of biomolecular structures, structure prediction performance has significantly improved, and the potential for biomedical applications has increased considerably. The prediction of protein-ligand complex structures, applicable to the atomistic understanding of biomolecular functions and the effective design of drug molecules, has also improved with the introduction of deep learning. In this paper, it is demonstrated that docking performance can be greatly enhanced by training an energy function that encapsulates physical effects using deep learning within the framework of the traditional protein-ligand docking method. The advantage of this method, called GalaxyDock-DL, lies in its minimal overfitting to the training data compared to several existing deep learning-based protein-ligand docking methods. Unlike some recent deep learning methods, it does not use information about known binding pocket center positions. Instead, the results of this docking method show a systematic dependence on the physical properties of the target protein-ligand complexes such as atomic thermal fluctuations and binding affinity. GalaxyDock-DL utilizes the global optimization technique of the conventional protein-ligand docking method, GalaxyDock, and a neural network energy function trained to stabilize the native state compared to non-native states, just as physical free energy does. This physical principle-based approach suggests directions not only for future structure prediction involving structurally flexible biomolecular complexes but also for predicting binding affinity, thereby providing guidance for the effective design of biofunctional ligands.

PMID:39109987 | DOI:10.1021/acs.jctc.4c00385

Eurasian J Med. 2023 Dec 29;55(1):91-97. doi: 10.5152/eurasianjmed.2023.23347.

ABSTRACT

The aging population challenges the health-care system with chronic diseases. Cerebrovascular diseases are important components of these chronic conditions. Stroke is the acute cessation of blood in the brain, which can lead to rapid tissue loss. Therefore, fast, accurate, and reliable automatic methods are required to facilitate stroke management. The performance of artificial intelligence (AI) methods is increasing in all domains. Vision tasks, including natural images and medical images, are particularly benefiting from the skills of AI models. The AI methods that can be applied to stroke imaging have a broad range, including classical machine learning tools such as support vector machines, random forests, logistic regression, and linear discriminant analysis, as well as deep learning models, such as convolutional neural networks, recurrent neural networks, autoencoders, and U-Net. Both tools can be applied to various aspects of stroke management, including time-to-event onset determination, stroke confirmation, large vessel occlusion detection, difusion restriction, perfusion deficit, core and penumbra identification, afected region segmentation, and functional outcome prediction. While building these AI models, maximum care should be exercised in order to reduce bias and build generalizable models. One of the most important prerequisites for building unbiased models is collecting large, diverse, and quality data that reflects the underlying population well and splitting the training and testing parts in a way that both represent a similar distribution. Explainability and trustworthiness are other important properties of machine learning models that could be widely adopted in clinical practices.

PMID:39109827 | DOI:10.5152/eurasianjmed.2023.23347

Global Spine J. 2024 Aug 7:21925682241270036. doi: 10.1177/21925682241270036. Online ahead of print.

ABSTRACT

STUDY DESIGN: Cross-sectional study.

OBJECTIVES: Imaging classification of adolescent idiopathic scoliosis (AIS) is directly related to the surgical strategy, but the artificial classification is complex and depends on doctors' experience. This study investigated deep learning-based automated classification methods (DL group) for AIS and validated the consistency of machine classification and manual classification (M group).

METHODS: A total of 506 cases (81 males and 425 females) and 1812 AIS full spine images in the anteroposterior (AP), lateral (LAT), left bending (LB) and right bending (RB) positions were retrospectively used for training. The mean age was 13.6 ± 1.8. The mean maximum Cobb angle was 46.8 ± 12.0. U-Net semantic segmentation neural network technology and deep learning methods were used to automatically segment and establish the alignment relationship between multiple views of the spine, and to extract spinal features such as the Cobb angle. The type of each test case was automatically calculated according to Lenke's rule. An additional 107 cases of adolescent idiopathic scoliosis imaging were prospectively used for testing. The consistency of the DL group and M group was compared.

RESULTS: Automatic vertebral body segmentation and recognition, multi-view alignment of the spine and automatic Cobb angle measurement were implemented. Compare to the M group, the consistency of the DL group was significantly higher in 3 aspects: type of lateral convexity (0.989 vs 0.566), lumbar curvature modifier (0.932 vs 0.738), and sagittal plane modifier (0.987 vs 0.522).

CONCLUSIONS: Deep learning enables automated Cobb angle measurement and automated Lenke classification of idiopathic scoliosis whole spine radiographs with higher consistency than manual measurement classification.

PMID:39109794 | DOI:10.1177/21925682241270036

J Comput Biol. 2024 Aug 7. doi: 10.1089/cmb.2024.0587. Online ahead of print.

ABSTRACT

Small molecules (SMs) play a pivotal role in regulating microRNAs (miRNAs). Existing prediction methods for associations between SM-miRNA have overlooked crucial aspects: the incorporation of local topological features between nodes, which represent either SMs or miRNAs, and the effective fusion of node features with topological features. This study introduces a novel approach, termed high-order topological features for SM-miRNA association prediction (HTFSMMA), which specifically addresses these limitations. Initially, an association graph is formed by integrating SM-miRNA association data, SM similarity, and miRNA similarity. Subsequently, we focus on the local information of links and propose target neighborhood graph convolutional network for extracting local topological features. Then, HTFSMMA employs graph attention networks to amalgamate these local features, thereby establishing a platform for the acquisition of high-order features through random walks. Finally, the extracted features are integrated into the multilayer perceptron to derive the association prediction scores. To demonstrate the performance of HTFSMMA, we conducted comprehensive evaluations including five-fold cross-validation, leave-one-out cross-validation (LOOCV), SM-fixed local LOOCV, and miRNA-fixed local LOOCV. The area under receiver operating characteristic curve values were 0.9958 ± 0.0024 (0.8722 ± 0.0021), 0.9986 (0.9504), 0.9974 (0.9111), and 0.9977 (0.9074), respectively. Our findings demonstrate the superior performance of HTFSMMA over existing approaches. In addition, three case studies and the DeLong test have confirmed the effectiveness of the proposed method. These results collectively underscore the significance of HTFSMMA in facilitating the inference of associations between SMs and miRNAs.

PMID:39109562 | DOI:10.1089/cmb.2024.0587

J Big Data. 2024;11(1):104. doi: 10.1186/s40537-024-00974-x. Epub 2024 Aug 4.

ABSTRACT

The morphology and distribution of airway tree abnormalities enable diagnosis and disease characterisation across a variety of chronic respiratory conditions. In this regard, airway segmentation plays a critical role in the production of the outline of the entire airway tree to enable estimation of disease extent and severity. Furthermore, the segmentation of a complete airway tree is challenging as the intensity, scale/size and shape of airway segments and their walls change across generations. The existing classical techniques either provide an undersegmented or oversegmented airway tree, and manual intervention is required for optimal airway tree segmentation. The recent development of deep learning methods provides a fully automatic way of segmenting airway trees; however, these methods usually require high GPU memory usage and are difficult to implement in low computational resource environments. Therefore, in this study, we propose a data-centric deep learning technique with big interpolated data, Interpolation-Split, to boost the segmentation performance of the airway tree. The proposed technique utilises interpolation and image split to improve data usefulness and quality. Then, an ensemble learning strategy is implemented to aggregate the segmented airway segments at different scales. In terms of average segmentation performance (dice similarity coefficient, DSC), our method (A) achieves 90.55%, 89.52%, and 85.80%; (B) outperforms the baseline models by 2.89%, 3.86%, and 3.87% on average; and (C) produces maximum segmentation performance gain by 14.11%, 9.28%, and 12.70% for individual cases when (1) nnU-Net with instant normalisation and leaky ReLU; (2) nnU-Net with batch normalisation and ReLU; and (3) modified dilated U-Net are used respectively. Our proposed method outperformed the state-of-the-art airway segmentation approaches. Furthermore, our proposed technique has low RAM and GPU memory usage, and it is GPU memory-efficient and highly flexible, enabling it to be deployed on any 2D deep learning model.

PMID:39109339 | PMC:PMC11298507 | DOI:10.1186/s40537-024-00974-x

Front Artif Intell. 2024 Jul 23;7:1363531. doi: 10.3389/frai.2024.1363531. eCollection 2024.

ABSTRACT

Deep learning models have achieved state-of-the-art performance for text classification in the last two decades. However, this has come at the expense of models becoming less understandable, limiting their application scope in high-stakes domains. The increased interest in explainability has resulted in many proposed forms of explanation. Nevertheless, recent studies have shown that rationales, or language explanations, are more intuitive and human-understandable, especially for non-technical stakeholders. This survey provides an overview of the progress the community has achieved thus far in rationalization approaches for text classification. We first describe and compare techniques for producing extractive and abstractive rationales. Next, we present various rationale-annotated data sets that facilitate the training and evaluation of rationalization models. Then, we detail proxy-based and human-grounded metrics to evaluate machine-generated rationales. Finally, we outline current challenges and encourage directions for future work.

PMID:39109323 | PMC:PMC11300430 | DOI:10.3389/frai.2024.1363531

Front Med (Lausanne). 2024 Jul 22;11:1391184. doi: 10.3389/fmed.2024.1391184. eCollection 2024.

ABSTRACT

INTRODUCTION: Tuberculosis (TB) stands as a paramount global health concern, contributing significantly to worldwide mortality rates. Effective containment of TB requires deployment of cost-efficient screening method with limited resources. To enhance the precision of resource allocation in the global fight against TB, this research proposed chest X-ray radiography (CXR) based machine learning screening algorithms with optimization, benchmarking and tuning for the best TB subclassification tasks for clinical application.

METHODS: This investigation delves into the development and evaluation of a robust ensemble deep learning framework, comprising 43 distinct models, tailored for the identification of active TB cases and the categorization of their clinical subtypes. The proposed framework is essentially an ensemble model with multiple feature extractors and one of three fusion strategies-voting, attention-based, or concatenation methods-in the fusion stage before a final classification. The comprised de-identified dataset contains records of 915 active TB patients alongside 1,276 healthy controls with subtype-specific information. Thus, the realizations of our framework are capable for diagnosis with subclass identification. The subclass tags include: secondary tuberculosis/tuberculous pleurisy; non-cavity/cavity; secondary tuberculosis only/secondary tuberculosis and tuberculous pleurisy; tuberculous pleurisy only/secondary tuberculosis and tuberculous pleurisy.

RESULTS: Based on the dataset and model selection and tuning, ensemble models show their capability with self-correction capability of subclass identification with rendering robust clinical predictions. The best double-CNN-extractor model with concatenation/attention fusion strategies may potentially be the successful model for subclass tasks in real application. With visualization techniques, in-depth analysis of the ensemble model's performance across different fusion strategies are verified.

DISCUSSION: The findings underscore the potential of such ensemble approaches in augmenting TB diagnostics with subclassification. Even with limited dataset, the self-correction within the ensemble models still guarantees the accuracies to some level for potential clinical decision-making processes in TB management. Ultimately, this study shows a direction for better TB screening in the future TB response strategy.

PMID:39109222 | PMC:PMC11301748 | DOI:10.3389/fmed.2024.1391184

iScience. 2024 Jun 25;27(7):110175. doi: 10.1016/j.isci.2024.110175. eCollection 2024 Jul 19.

ABSTRACT

Accurate geographical traffic forecasting plays a critical role in urban transportation planning, traffic management, and geospatial artificial intelligence (GeoAI). Although deep learning models have made significant progress in geographical traffic forecasting, they still face challenges in effectively capturing long-term temporal dependencies and modeling heterogeneous dynamic spatial dependencies. To address these issues, we propose a novel deep transformer-based heterogeneous spatiotemporal graph learning model for geographical traffic forecasting. Our model incorporates a temporal transformer that captures long-term temporal patterns in traffic data without simple data fusion. Furthermore, we introduce adaptive normalized graph structures within different graph layers, enabling the model to capture dynamic spatial dependencies and adapt to diverse traffic scenarios, especially for the heterogeneous relationship. We conduct comprehensive experiments and visualization on four primary public datasets and demonstrate that our model achieves state-of-the-art results in comparison to existing methods.

PMID:39109176 | PMC:PMC11302005 | DOI:10.1016/j.isci.2024.110175