NPJ Precis Oncol. 2024 Mar 5;8(1):56. doi: 10.1038/s41698-024-00539-4.

ABSTRACT

Considering the profound transformation affecting pathology practice, we aimed to develop a scalable artificial intelligence (AI) system to diagnose colorectal cancer from whole-slide images (WSI). For this, we propose a deep learning (DL) system that learns from weak labels, a sampling strategy that reduces the number of training samples by a factor of six without compromising performance, an approach to leverage a small subset of fully annotated samples, and a prototype with explainable predictions, active learning features and parallelisation. Noting some problems in the literature, this study is conducted with one of the largest WSI colorectal samples dataset with approximately 10,500 WSIs. Of these samples, 900 are testing samples. Furthermore, the robustness of the proposed method is assessed with two additional external datasets (TCGA and PAIP) and a dataset of samples collected directly from the proposed prototype. Our proposed method predicts, for the patch-based tiles, a class based on the severity of the dysplasia and uses that information to classify the whole slide. It is trained with an interpretable mixed-supervision scheme to leverage the domain knowledge introduced by pathologists through spatial annotations. The mixed-supervision scheme allowed for an intelligent sampling strategy effectively evaluated in several different scenarios without compromising the performance. On the internal dataset, the method shows an accuracy of 93.44% and a sensitivity between positive (low-grade and high-grade dysplasia) and non-neoplastic samples of 0.996. On the external test samples varied with TCGA being the most challenging dataset with an overall accuracy of 84.91% and a sensitivity of 0.996.

PMID:38443695 | DOI:10.1038/s41698-024-00539-4

Nat Cancer. 2024 Mar 5. doi: 10.1038/s43018-024-00740-1. Online ahead of print.

ABSTRACT

Cyclin-dependent kinase 4 and 6 inhibitors (CDK4/6is) have revolutionized breast cancer therapy. However, <50% of patients have an objective response, and nearly all patients develop resistance during therapy. To elucidate the underlying mechanisms, we constructed an interpretable deep learning model of the response to palbociclib, a CDK4/6i, based on a reference map of multiprotein assemblies in cancer. The model identifies eight core assemblies that integrate rare and common alterations across 90 genes to stratify palbociclib-sensitive versus palbociclib-resistant cell lines. Predictions translate to patients and patient-derived xenografts, whereas single-gene biomarkers do not. Most predictive assemblies can be shown by CRISPR-Cas9 genetic disruption to regulate the CDK4/6i response. Validated assemblies relate to cell-cycle control, growth factor signaling and a histone regulatory complex that we show promotes S-phase entry through the activation of the histone modifiers KAT6A and TBL1XR1 and the transcription factor RUNX1. This study enables an integrated assessment of how a tumor's genetic profile modulates CDK4/6i resistance.

PMID:38443662 | DOI:10.1038/s43018-024-00740-1

Sci Rep. 2024 Mar 5;14(1):5416. doi: 10.1038/s41598-024-55990-x.

ABSTRACT

The presence of light absorption and scattering in underwater conditions results in underwater images with missing details, low contrast, and color bias. The current deep learning-based methods bring unlimited potential for underwater image restoration (UIR) tasks. These methods, however, do not adequately take into account the inconsistency of the attenuation of different color channels and spatial regions when performing image restoration. To solve these gaps, we propose a dual information modulation network (DIMN) for accurate UIR tasks. To be specific, we design a multi-information enhancement module (MIEM), empowered by spatial-aware attention block (SAAB) and multi-scale structural Transformer block (MSTB), to guide the inductive bias of image degradation processes under nonhomogeneous media distributions. SAAB focuses on different spatial locations, capturing more spatial-aware cues to correct color deviations and recover details. MSTB utilizes the difference and complementarity between features at different scales to effectively complement the network's structural and global perceptual capabilities, enhancing image sharpness and contrast further. Experimental results reveal that the proposed DIMN exceeds most state-of-the-art UIR methods. Our code and results are available at: https://github.com/wwaannggllii/DIMN .

PMID:38443587 | DOI:10.1038/s41598-024-55990-x

Sci Rep. 2024 Mar 5;14(1):5447. doi: 10.1038/s41598-024-56066-6.

ABSTRACT

In recent years, automatic image classification methods have significantly progressed, notably black box algorithms such as machine learning and deep learning. Unfortunately, such efforts only focused on improving performance, rather than attempting to explain and interpret how classification models actually operate. This article compares three state-of-the-art algorithms incorporating random forests, gradient boosting and convolutional neural networks for geomorphological mapping. It also attempts to explain how the most effective classifier makes decisions by evaluating which of the geomorphometric variables are most important for automatic mapping and how they affect the classification results using one of the explainable artificial intelligence techniques, namely accumulated local effects (ALE). This method allows us to understand the relationship between predictors and the model's outcome. For these purposes, eight sheets of the digital geomorphological map of Poland on the scale of 1:100,000 were used as the reference material. The classification results were validated using the holdout method and cross-validation for individual sheets representing different morphogenetic zones. The terrain elevation entropy, absolute elevation, aggregated median elevation and standard deviation of elevation had the greatest impact on the classification results among the 15 geomorphometric variables considered. The ALE analysis was conducted for the XGBoost classifier, which achieved the highest accuracy of 92.8%, ahead of Random Forests at 84% and LightGBM at 73.7% and U-Net at 59.8%. We conclude that automatic classification can support geomorphological mapping only if the geomorphological characteristics in the predicted area are similar to those in the training dataset. The ALE plots allow us to analyze the relationship between geomorphometric variables and landform membership, which helps clarify their role in the classification process.

PMID:38443550 | DOI:10.1038/s41598-024-56066-6

Sci Rep. 2024 Mar 5;14(1):5383. doi: 10.1038/s41598-024-54048-2.

ABSTRACT

Breast density, or the amount of fibroglandular tissue (FGT) relative to the overall breast volume, increases the risk of developing breast cancer. Although previous studies have utilized deep learning to assess breast density, the limited public availability of data and quantitative tools hinders the development of better assessment tools. Our objective was to (1) create and share a large dataset of pixel-wise annotations according to well-defined criteria, and (2) develop, evaluate, and share an automated segmentation method for breast, FGT, and blood vessels using convolutional neural networks. We used the Duke Breast Cancer MRI dataset to randomly select 100 MRI studies and manually annotated the breast, FGT, and blood vessels for each study. Model performance was evaluated using the Dice similarity coefficient (DSC). The model achieved DSC values of 0.92 for breast, 0.86 for FGT, and 0.65 for blood vessels on the test set. The correlation between our model's predicted breast density and the manually generated masks was 0.95. The correlation between the predicted breast density and qualitative radiologist assessment was 0.75. Our automated models can accurately segment breast, FGT, and blood vessels using pre-contrast breast MRI data. The data and the models were made publicly available.

PMID:38443410 | DOI:10.1038/s41598-024-54048-2

Nat Commun. 2024 Mar 5;15(1):2011. doi: 10.1038/s41467-024-46241-8.

ABSTRACT

Translation elongation is essential for maintaining cellular proteostasis, and alterations in the translational landscape are associated with a range of diseases. Ribosome profiling allows detailed measurements of translation at the genome scale. However, it remains unclear how to disentangle biological variations from technical artifacts in these data and identify sequence determinants of translation dysregulation. Here we present Riboformer, a deep learning-based framework for modeling context-dependent changes in translation dynamics. Riboformer leverages the transformer architecture to accurately predict ribosome densities at codon resolution. When trained on an unbiased dataset, Riboformer corrects experimental artifacts in previously unseen datasets, which reveals subtle differences in synonymous codon translation and uncovers a bottleneck in translation elongation. Further, we show that Riboformer can be combined with in silico mutagenesis to identify sequence motifs that contribute to ribosome stalling across various biological contexts, including aging and viral infection. Our tool offers a context-aware and interpretable approach for standardizing ribosome profiling datasets and elucidating the regulatory basis of translation kinetics.

PMID:38443396 | DOI:10.1038/s41467-024-46241-8

ACS Sens. 2024 Mar 5. doi: 10.1021/acssensors.3c02789. Online ahead of print.

ABSTRACT

Extracellular vesicle microRNAs (EV miRNAs) are critical noninvasive biomarkers for early cancer diagnosis. However, accurate cancer diagnosis based on bulk analysis is hindered by the heterogeneity among EVs. Herein, we report an approach for profiling single-EV multi-miRNA signatures by combining total internal reflection fluorescence (TIRF) imaging with a deep learning (DL) algorithm for the first time. This innovative technique allows for the precise characterization of EV miRNAs at the single-vesicle level, overcoming the challenges posed by EV heterogeneity. TIRF with high resolution and a signal-to-noise ratio can simultaneously detect multi-miRNAs in situ in individual EVs. DL algorithm avoids complicated and inaccurate artificial feature extraction, achieving automated high-resolution image analysis. Using this approach, we reveal that the main variation of EVs from 5 cancer cells and normal plasma is the triple-positive EV subpopulation, and the classification accuracy of single triple-positive EVs from 6 sources can reach above 95%. In the clinical cohort, 20 patients (5 lung cancer, 5 breast cancer, 5 cervical cancer, and 5 colon cancer) and 5 healthy controls are predicted with an overall accuracy of 100%. This single-EV strategy provides new opportunities for exploring more specific EV biomarkers to achieve cancer diagnosis and classification.

PMID:38442411 | DOI:10.1021/acssensors.3c02789

J Phys Chem Lett. 2024 Mar 5:2841-2850. doi: 10.1021/acs.jpclett.4c00100. Online ahead of print.

ABSTRACT

Deep learning models have been widely used for high-performance material property prediction. However, training such models usually requires a large amount of labeled data, which are usually unavailable. Self-supervised learning (SSL) methods have been proposed to address this data scarcity issue. Herein, we present DSSL, a physics-guided dual SSL framework, for graph neural network-based material property prediction, which combines node masking-based generative SSL with atomic coordinate perturbation-based contrastive SSL strategies to capture local and global information about input crystals. Moreover, we achieve physics-guided pretraining by using the macroproperty (e.g., elasticity)-related microproperty prediction of atomic stiffness as an additional pretext task. We pretrain our DSSL model on the Materials Project database and fine-tune it with 10 material property data sets. The experimental results demonstrate that teaching neural networks some physics using the SSL strategy can afford ≤26.89% performance improvement compared to that of the baseline models.

PMID:38442260 | DOI:10.1021/acs.jpclett.4c00100

Chaos. 2024 Mar 1;34(3):033108. doi: 10.1063/5.0187866.

ABSTRACT

Nonlinear dynamical systems with control parameters may not be well modeled by shallow neural networks. In this paper, the stable fixed-point solutions, periodic and chaotic solutions of the parameter-dependent Lorenz system are learned simultaneously via a very deep neural network. The proposed deep learning model consists of a large number of identical linear layers, which provide excellent nonlinear mapping capability. Residual connections are applied to ease the flow of information and a large training dataset is further utilized. Extensive numerical results show that the chaotic solutions can be accurately forecasted for several Lyapunov times and long-term predictions are achieved for periodic solutions. Additionally, the dynamical characteristics such as bifurcation diagrams and largest Lyapunov exponents can be well recovered from the learned solutions. Finally, the principal factors contributing to the high prediction accuracy are discussed.

PMID:38442234 | DOI:10.1063/5.0187866

PLoS One. 2024 Mar 5;19(3):e0298160. doi: 10.1371/journal.pone.0298160. eCollection 2024.

ABSTRACT

Contrails are line-shaped clouds formed in the exhaust of aircraft engines that significantly contribute to global warming. This paper confidently proposes integrating advanced image segmentation techniques to identify and monitor aircraft contrails to address the challenges associated with climate change. We propose the SegX-Net architecture, a highly efficient and lightweight model that combines the DeepLabV3+, upgraded, and ResNet-101 architectures to achieve superior segmentation accuracy. We evaluated the performance of our model on a comprehensive dataset from Google research and rigorously measured its efficacy with metrics such as IoU, F1 score, Sensitivity and Dice Coefficient. Our results demonstrate that our enhancements have significantly improved the efficacy of the SegX-Net model, with an outstanding IoU score of 98.86% and an impressive F1 score of 99.47%. These results unequivocally demonstrate the potential of image segmentation methods to effectively address and mitigate the impact of air conflict on global warming. Using our proposed SegX-Net architecture, stakeholders in the aviation industry can confidently monitor and mitigate the impact of aircraft shrinkage on the environment, significantly contributing to the global fight against climate change.

PMID:38442105 | DOI:10.1371/journal.pone.0298160

PLoS One. 2024 Mar 5;19(3):e0276155. doi: 10.1371/journal.pone.0276155. eCollection 2024.

ABSTRACT

Water quality prediction is of great significance in pollution control, prevention, and management. Deep learning models have been applied to water quality prediction in many recent studies. However, most existing deep learning models for water quality prediction are used for single-site data, only considering the time dependency of water quality data and ignoring the spatial correlation among multi-sites. This research defines and analyzes the non-aligned spatial correlations that exist in multi-site water quality data. Then deploy spatial-temporal graph convolution to process water quality data, which takes into account both the temporal and spatial correlation of multi-site water quality data. A multi-site water pollution prediction method called W-WaveNet is proposed that integrates adaptive graph convolution and Convolutional Neural Network, Long Short-Term Memory (CNN-LSTM). It integrates temporal and spatial models by interleaved stacking. Theoretical analysis shows that the method can deal with non-aligned spatial correlations in different time spans, which is suitable for water quality data processing. The model validates water quality data generated on two real river sections that have multiple sites. The experimental results were compared with the results of Support Vector Regression, CNN-LSTM, and Spatial-Temporal Graph Convolutional Networks (STGCN). It shows that when W-WaveNet predicts water quality over two river sections, the average Mean Absolute Error is 0.264, which is 45.2% lower than the commonly used CNN-LSTM model and 23.8% lower than the STGCN. The comparison experiments also demonstrate that W-WaveNet has a more stable performance in predicting longer sequences.

PMID:38442101 | DOI:10.1371/journal.pone.0276155

IEEE Trans Image Process. 2024 Mar 5;PP. doi: 10.1109/TIP.2024.3371351. Online ahead of print.

ABSTRACT

Deep compressive sensing (CS) has become a prevalent technique for image acquisition and reconstruction. However, existing deep learning (DL)-based CS methods often encounter challenges such as block artifacts and information loss during iterative reconstruction, particularly at low sampling rates, resulting in a reduction of reconstructed details. To address these issues, we propose NesTD-Net, an unfolding-based architecture inspired by the NESTA algorithm, designed for image CS. NesTD-Net integrates DL modules into NESTA iterations, forming a deep network that continuously iterates to minimize the ℓ1-norm CS problem, ensuring high-quality image CS. Utilizing a learned sampling matrix for measurements and an initialization module for initial estimate, NesTD-Net then introduces Iteration Sub-Modules derived from the NESTA algorithm (i.e., Yk, Zk, and Xk) during reconstruction stages to iteratively solve the ℓ1-norm CS reconstruction. Additionally, NesTD-Net incorporates a Dual-Path Deblocking Structure (DPDS) to facilitate feature information flow and mitigate block artifacts, enhancing image detail reconstruction. Furthermore, DPDS exhibits remarkable versatility and demonstrates seamless integration with other unfolding-based methods, offering the potential to enhance their performance in image reconstruction. Experimental results demonstrate that our proposed NesTD-Net achieves better performance compared to other state-of-the-art methods in terms of image quality metrics such as SSIM and PSNR, as well as visual perception on several public benchmark datasets. Our code is available at NesTD-Net.

PMID:38442062 | DOI:10.1109/TIP.2024.3371351

IEEE J Biomed Health Inform. 2024 Mar 5;PP. doi: 10.1109/JBHI.2024.3373124. Online ahead of print.

ABSTRACT

Cardiac valve event timing plays a crucial role when conducting clinical measurements using echocardiography. However, established automated approaches are limited by the need of external electrocardiogram sensors, and manual measurements often rely on timing from different cardiac cycles. Recent methods have applied deep learning to cardiac timing, but they have mainly been restricted to only detecting two key time points, namely end-diastole (ED) and end-systole (ES). In this work, we propose a deep learning approach that leverages triplane recordings to enhance detection of valve events in echocardiography. Our method demonstrates improved performance detecting six different events, including valve events conventionally associated with ED and ES. Of all events, we achieve an average absolute frame difference (aFD) of maximum 1.4 frames (29 ms) for start of diastasis, down to 0.6 frames (12 ms) for mitral valve opening when performing a ten-fold cross-validation with test splits on triplane data from 240 patients. On an external independent test consisting of apical long-axis data from 180 other patients, the worst performing event detection had an aFD of 1.8 (30 ms). The proposed approach has the potential to significantly impact clinical practice by enabling more accurate, rapid and comprehensive event detection, leading to improved clinical measurements.

PMID:38442058 | DOI:10.1109/JBHI.2024.3373124

IEEE J Biomed Health Inform. 2024 Mar 5;PP. doi: 10.1109/JBHI.2024.3368500. Online ahead of print.

ABSTRACT

The integration of structural magnetic resonance imaging (sMRI) and deep learning techniques is one of the important research directions for the automatic diagnosis of Alzheimer's disease (AD). Despite the satisfactory performance achieved by existing voxel-based models based on convolutional neural networks (CNNs), such models only handle AD-related brain atrophy at a single spatial scale and lack spatial localization of abnormal brain regions based on model interpretability. To address the above limitations, we propose a traceable interpretability model for AD recognition based on multi-patch attention (MAD-Former). MAD-Former consists of two parts: recognition and interpretability. In the recognition part, we design a 3D brain feature extraction network to extract local features, followed by constructing a dual-branch attention structure with different patch sizes to achieve global feature extraction, forming a multi-scale spatial feature extraction framework. Meanwhile, we propose an important attention similarity position loss function to assist in model decision-making. The interpretability part proposes a traceable method that can obtain a 3D ROI space through attention-based selection and receptive field tracing. This space encompasses key brain tissues that influence model decisions. Experimental results reveal the significant role of brain tissues such as the Fusiform Gyrus (FuG) in AD recognition. MAD-Former achieves outstanding performance in different tasks on ADNI and OASIS datasets, demonstrating reliable model interpretability.

PMID:38442047 | DOI:10.1109/JBHI.2024.3368500

J Invasive Cardiol. 2024 Mar;36(3). doi: 10.25270/jic/23.00285.

ABSTRACT

OBJECTIVES: Coronary angiography (CAG)-derived physiology methods have been developed in an attempt to simplify and increase the usage of coronary physiology, based mostly on dynamic fluid computational algorithms. We aimed to develop a different approach based on artificial intelligence methods, which has seldom been explored.

METHODS: Consecutive patients undergoing invasive instantaneous free-wave ratio (iFR) measurements were included. We developed artificial intelligence (AI) models capable of classifying target lesions as positive (iFR ≤ 0.89) or negative (iFR > 0.89). The predictions were then compared to the true measurements.

RESULTS: Two hundred-fifty measurements were included, and 3 models were developed. Model 3 had the best overall performance: accuracy, negative predictive value (NPV), positive predictive value (PPV), sensitivity, and specificity were 69%, 88%, 44%, 74%, and 67%, respectively. Performance differed per target vessel. For the left anterior descending artery (LAD), model 3 had the highest accuracy (66%), while model 2 the highest NPV (86%) and sensitivity (91%). PPV was always low/modest. Model 1 had the highest specificity (68%). For the right coronary artery, model 1's accuracy was 86%, NPV was 97%, and specificity was 87%, but all models had low PPV (maximum 25%) and low/modest sensitivity (maximum 60%). For the circumflex, model 1 performed best: accuracy, NPV, PPV, sensitivity, and specificity were 69%, 96%, 24%, 80%, and 68%, respectively.

CONCLUSIONS: We developed 3 AI models capable of binary iFR estimation from CAG images. Despite modest accuracy, the consistently high NPV is of potential clinical significance, as it would enable avoiding further invasive maneuvers after CAG. This pivotal study offers proof of concept for further development.

PMID:38441988 | DOI:10.25270/jic/23.00285

J Chem Theory Comput. 2024 Mar 5. doi: 10.1021/acs.jctc.3c01415. Online ahead of print.

ABSTRACT

Simultaneous prediction of the molecular response properties, such as polarizability and the NMR shielding constant, at a low computational cost is an unresolved issue. We propose to combine a linear-scaling generalized energy-based fragmentation (GEBF) method and deep learning (DL) with both molecular and atomic information-theoretic approach (ITA) quantities as effective descriptors. In GEBF, the total molecular polarizability can be assembled as a linear combination of the corresponding quantities calculated from a set of small embedded subsystems in GEBF. In the new GEBF-DL(ITA) protocol, one can predict subsystem polarizabilities based on the corresponding molecular wave function (thus electron density and ITA quantities) and DL model rather than calculate them from the computationally intensive coupled-perturbed Hartree-Fock or Kohn-Sham equations and finally obtain the total molecular polarizability via a linear combination equation. As a proof-of-concept application, we predict the molecular polarizabilities of large proteins and protein aggregates. GEBF-DL(ITA) is shown to be as accurate enough as GEBF, with mean absolute percentage error <1%. For the largest protein aggregate (>4000 atoms), GEBF-DL(ITA) gains a speedup ratio of 3 compared with GEBF. It is anticipated that when more advanced electronic structure methods are used, this advantage will be more appealing. Moreover, one can also predict the NMR chemical shieldings of proteins with reasonably good accuracy. Overall, the cost-efficient GEBF-DL(ITA) protocol should be a robust theoretical tool for simultaneously predicting polarizabilities and NMR shieldings of large systems.

PMID:38441881 | DOI:10.1021/acs.jctc.3c01415

J Chem Inf Model. 2024 Mar 5. doi: 10.1021/acs.jcim.3c01961. Online ahead of print.

ABSTRACT

Improving the generalization ability of scoring functions remains a major challenge in protein-ligand binding affinity prediction. Many machine learning methods are limited by their reliance on single-modal representations, hindering a comprehensive understanding of protein-ligand interactions. We introduce a graph-neural-network-based scoring function that utilizes a triplet contrastive learning loss to improve protein-ligand representations. In this model, three-dimensional complex representations and the fusion of two-dimensional ligand and coarse-grained pocket representations converge while distancing from decoy representations in latent space. After rigorous validation on multiple external data sets, our model exhibits commendable generalization capabilities compared to those of other deep learning-based scoring functions, marking it as a promising tool in the realm of drug discovery. In the future, our training framework can be extended to other biophysical- and biochemical-related problems such as protein-protein interaction and protein mutation prediction.

PMID:38441880 | DOI:10.1021/acs.jcim.3c01961

Ophthalmol Ther. 2024 Mar 5. doi: 10.1007/s40123-024-00908-z. Online ahead of print.

ABSTRACT

INTRODUCTION: This study aims to quantitatively assess diffuse chorioretinal atrophy (DCA) in pathologic myopia and establish a standardized classification system utilizing artificial intelligence.

METHODS: A total of 202 patients underwent comprehensive examinations, and 338 eyes were included in the study. The methodology involved image preprocessing, sample labeling, employing deep learning segmentation models, measuring and calculating the area and density of DCA lesions. Lesion severity of DCA was graded using statistical methods, and grades were assigned to describe the morphology of corresponding fundus photographs. Hierarchical clustering was employed to categorize diffuse atrophy fundus into three groups based on the area and density of diffuse atrophy (G1, G2, G3), while high myopic fundus without diffuse atrophy was designated as G0. One-way analysis of variance (ANOVA) and nonparametric tests were conducted to assess the statistical association with different grades of DCA.

RESULTS: On the basis of the area and density of DCA, the condition was classified into four grades: G0, G1 (0 < density ≤ 0.093), G2 (0.093 < density ≤ 0.245), and G3 (0.245 < density ≤ 0.712). Fundus photographs depicted a progressive enlargement of atrophic lesions, evolving from punctate-shaped to patchy with indistinct boundaries. DCA atrophy lesions exhibited a gradual shift in color from brown-yellow to yellow-white, originating from the temporal side of the optic disc and extending towards the macula, with severe cases exhibiting widespread distribution throughout the posterior pole. Patients with DCA were significantly older [34.00 (27.00, 48.00) vs 29.00 (26.00, 34.00) years], possessed a longer axial length (28.85 ± 1.57 vs 27.11 ± 1.01 mm), and exhibited a more myopic spherical equivalent [- 13.00 (- 16.00, - 10.50) vs - 9.09 ± 2.41 D] compared to those without DCA (G0) (all P < 0.001). In eyes with DCA, a trend emerged as grades increased from G1 to G3, showing associations with older age, longer axial length, deeper myopic spherical equivalent, larger area of parapapillary atrophy, and increased fundus tessellated density (all P < 0.001).

CONCLUSIONS: The novel grading system for DCA, based on assessments of area and density, serves as a reliable measure for evaluating the severity of this condition, making it suitable for widespread application in the screening of pathologic myopia.

PMID:38441856 | DOI:10.1007/s40123-024-00908-z

J Imaging Inform Med. 2024 Mar 5. doi: 10.1007/s10278-024-01063-4. Online ahead of print.

ABSTRACT

This study aimed to develop a model based on radiomics and deep learning features to predict the ablation rate in patients with adenomyosis undergoing high-intensity focused ultrasound (HIFU) therapy. A total of 119 patients with adenomyosis who received HIFU therapy were retrospectively analyzed. Participants were included in the training and testing queues in a 7:3 ratio. Radiomics features were extracted from T2-weighted imaging (T2WI) images, and VGG-19 was used to extract advanced deep features. An ensemble model based on multi-model fusion for predicting the efficacy of HIFU in adenomyosis was proposed, which consists of four base classifiers and was evaluated using accuracy, precision, recall, F-score, and area under the receiver operating characteristic curve (AUC). The predictive performance of the combined model combining radiomics and deep learning features outperformed the radiomics and deep learning feature models alone, with accuracy of 0.848 and 0.814 in training and test sets, and AUC of 0.916 and 0.861, respectively. Compared with the base classifiers that make up the multi-model fusion model, the fusion model also exhibited better prediction performance. The fusion model incorporating both radiomics and deep learning features had certain predictive value for the ablation rate of adenomyosis under HIFU therapy and could help select patients with adenomyosis who would benefit from HIFU therapy.

PMID:38441701 | DOI:10.1007/s10278-024-01063-4

J Imaging Inform Med. 2024 Mar 5. doi: 10.1007/s10278-024-01061-6. Online ahead of print.

ABSTRACT

The utilization of advanced intraoral scanners to acquire 3D dental models has gained significant popularity in the fields of dentistry and orthodontics. Accurate segmentation and labeling of teeth on digitized 3D dental surface models are crucial for computer-aided treatment planning. At the same time, manual labeling of these models is a time-consuming task. Recent advances in geometric deep learning have demonstrated remarkable efficiency in surface segmentation when applied to raw 3D models. However, segmentation of the dental surface remains challenging due to the atypical and diverse appearance of the patients' teeth. Numerous deep learning methods have been proposed to automate dental surface segmentation. Nevertheless, they still show limitations, particularly in cases where teeth are missing or severely misaligned. To overcome these challenges, we introduce a network operator called dilated edge convolution, which enhances the network's ability to learn additional, more distant features by expanding its receptive field. This leads to improved segmentation results, particularly in complex and challenging cases. To validate the effectiveness of our proposed method, we performed extensive evaluations on the recently published benchmark data set for dental model segmentation Teeth3DS. We compared our approach with several other state-of-the-art methods using a quantitative and qualitative analysis. Through these evaluations, we demonstrate the superiority of our proposed method, showcasing its ability to outperform existing approaches in dental surface segmentation.

PMID:38441700 | DOI:10.1007/s10278-024-01061-6