J Hazard Mater. 2024 May 22;474:134651. doi: 10.1016/j.jhazmat.2024.134651. Online ahead of print.

ABSTRACT

As emerging pollutants, antidepressants (AD) must be urgently investigated for risk identification and assessment. This study constructed a comprehensive-effect risk-priority screening system (ADRank) for ADs by characterizing AD functionality, occurrence, persistence, bioaccumulation and toxicity based on the integrated assignment method. A classification model for ADs was constructed using an improved mixup-transformer deep learning method, and its classification accuracy was compared with those of other models. The accuracy of the proposed model improved by up to 23.25 % compared with the random forest model, and the reliability was 80 % more than that of the TOPSIS method. A priority screening candidate list was proposed to screen 33 high-priority ADs. Finally, SHapley Additive explanation (SHAP) visualization, molecular dynamics, and amino acid analysis were performed to analyze the correlation between AD structure and toxic receptor binding characteristics and reveal the differences in AD risk priority. ADs with more intramolecular hydrogen bonds, higher hydrophobicity, and electronegativity had a more significant risk. Van der Waals and electrostatic interactions were the primary influencing factors, and significant differences in the types and proportions of the main amino acids in the interaction between ADs and receptors were observed. The results of the study provide constructive schemes and insights for AD priority screening and risk management.

PMID:38843640 | DOI:10.1016/j.jhazmat.2024.134651

Comput Methods Programs Biomed. 2024 May 28;254:108252. doi: 10.1016/j.cmpb.2024.108252. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Hepatocellular carcinoma is a common disease with high mortality. Through deep learning methods to analyze HCC CT, the screening classification and prognosis model of HCC can be established, which further promotes the development of computer-aided diagnosis and treatment in the treatment of HCC. However, there are significant challenges in the actual establishment of HCC auxiliary diagnosis model due to data imbalance, especially for rare subtypes of HCC and underrepresented demographic groups. This study proposes a GAN model aimed at overcoming these obstacles and improving the accuracy of HCC auxiliary diagnosis.

METHODS: In order to generate liver and tumor images close to the real distribution. Firstly, we construct a new gradient transfer sampling module to improve the lack of texture details and excessive gradient transfer parameters of the deep model; Secondly, we construct an attention module with spatial and cross-channel feature extraction ability to improve the discriminator's ability to distinguish images; Finally, we design a new loss function for liver tumor imaging features to constrain the model to approach the real tumor features in iterations.

RESULTS: In qualitative analysis, the images synthetic by our method closely resemble the real images in liver parenchyma, blood vessels, tumors, and other parts. In quantitative analysis, the optimal results of FID, PSNR, and SSIM are 75.73, 22.77, and 0.74, respectively. Furthermore, our experiments establish classification models for imbalanced data and enhanced data, resulting in an increase in accuracy rate by 21%-34%, an increase in AUC by 0.29 - 0.33, and an increase in specificity to 0.89.

CONCLUSION: Our solution provides a variety of training data sources with low cost and high efficiency for the establishment of classification or prognostic models for imbalanced data.

PMID:38843572 | DOI:10.1016/j.cmpb.2024.108252

Comput Biol Med. 2024 Jun 4;178:108710. doi: 10.1016/j.compbiomed.2024.108710. Online ahead of print.

ABSTRACT

BACKGROUND: Efficient and precise diagnosis of non-small cell lung cancer (NSCLC) is quite critical for subsequent targeted therapy and immunotherapy. Since the advent of whole slide images (WSIs), the transition from traditional histopathology to digital pathology has aroused the application of convolutional neural networks (CNNs) in histopathological recognition and diagnosis. HookNet can make full use of macroscopic and microscopic information for pathological diagnosis, but it cannot integrate other excellent CNN structures. The new version of HookEfficientNet is based on a combination of HookNet structure and EfficientNet that performs well in the recognition of general objects. Here, a high-precision artificial intelligence-guided histopathological recognition system was established by HookEfficientNet to provide a basis for the intelligent differential diagnosis of NSCLC.

METHODS: A total of 216 WSIs of lung adenocarcinoma (LUAD) and 192 WSIs of lung squamous cell carcinoma (LUSC) were recruited from the First Affiliated Hospital of Zhengzhou University. Deep learning methods based on HookEfficientNet, HookNet and EfficientNet B4-B6 were developed and compared with each other using area under the curve (AUC) and the Youden index. Temperature scaling was used to calibrate the heatmap and highlight the cancer region of interest. Four pathologists of different levels blindly reviewed 108 WSIs of LUAD and LUSC, and the diagnostic results were compared with the various deep learning models.

RESULTS: The HookEfficientNet model outperformed HookNet and EfficientNet B4-B6. After temperature scaling, the HookEfficientNet model achieved AUCs of 0.973, 0.980, and 0.989 and Youden index values of 0.863, 0.899, and 0.922 for LUAD, LUSC and normal lung tissue, respectively, in the testing set. The accuracy of the model was better than the average accuracy from experienced pathologists, and the model was superior to pathologists in the diagnosis of LUSC.

CONCLUSIONS: HookEfficientNet can effectively recognize LUAD and LUSC with performance superior to that of senior pathologists, especially for LUSC. The model has great potential to facilitate the application of deep learning-assisted histopathological diagnosis for LUAD and LUSC in the future.

PMID:38843570 | DOI:10.1016/j.compbiomed.2024.108710

BMC Med Inform Decis Mak. 2024 Jun 6;24(1):159. doi: 10.1186/s12911-024-02564-6.

ABSTRACT

BACKGROUND: Compared with the time-consuming and labor-intensive for biological validation in vitro or in vivo, the computational models can provide high-quality and purposeful candidates in an instant. Existing computational models face limitations in effectively utilizing sparse local structural information for accurate predictions in circRNA-disease associations. This study addresses this challenge with a proposed method, CDA-DGRL (Prediction of CircRNA-Disease Association based on Double-line Graph Representation Learning), which employs a deep learning framework leveraging graph networks and a dual-line representation model integrating graph node features.

METHOD: CDA-DGRL comprises several key steps: initially, the integration of diverse biological information to compute integrated similarities among circRNAs and diseases, leading to the construction of a heterogeneous network specific to circRNA-disease associations. Subsequently, circRNA and disease node features are derived using sparse autoencoders. Thirdly, a graph convolutional neural network is employed to capture the local graph network structure by inputting the circRNA-disease heterogeneous network alongside node features. Fourthly, the utilization of node2vec facilitates depth-first sampling of the circRNA-disease heterogeneous network to grasp the global graph network structure, addressing issues associated with sparse raw data. Finally, the fusion of local and global graph network structures is inputted into an extra trees classifier to identify potential circRNA-disease associations.

RESULTS: The results, obtained through a rigorous five-fold cross-validation on the circR2Disease dataset, demonstrate the superiority of CDA-DGRL with an AUC value of 0.9866 and an AUPR value of 0.9897 compared to existing state-of-the-art models. Notably, the hyper-random tree classifier employed in this model outperforms other machine learning classifiers.

CONCLUSION: Thus, CDA-DGRL stands as a promising methodology for reliably identifying circRNA-disease associations, offering potential avenues to alleviate the necessity for extensive traditional biological experiments. The source code and data for this study are available at https://github.com/zywait/CDA-DGRL .

PMID:38844961 | DOI:10.1186/s12911-024-02564-6

J Imaging Inform Med. 2024 Jun 6. doi: 10.1007/s10278-024-01138-2. Online ahead of print.

ABSTRACT

Artificial intelligence-enhanced identification of organs, lesions, and other structures in medical imaging is typically done using convolutional neural networks (CNNs) designed to make voxel-accurate segmentations of the region of interest. However, the labels required to train these CNNs are time-consuming to generate and require attention from subject matter experts to ensure quality. For tasks where voxel-level precision is not required, object detection models offer a viable alternative that can reduce annotation effort. Despite this potential application, there are few options for general-purpose object detection frameworks available for 3-D medical imaging. We report on MedYOLO, a 3-D object detection framework using the one-shot detection method of the YOLO family of models and designed for use with medical imaging. We tested this model on four different datasets: BRaTS, LIDC, an abdominal organ Computed tomography (CT) dataset, and an ECG-gated heart CT dataset. We found our models achieve high performance on a diverse range of structures even without hyperparameter tuning, reaching mean average precision (mAP) at intersection over union (IoU) 0.5 of 0.861 on BRaTS, 0.715 on the abdominal CT dataset, and 0.995 on the heart CT dataset. However, the models struggle with some structures, failing to converge on LIDC resulting in a mAP@0.5 of 0.0.

PMID:38844717 | DOI:10.1007/s10278-024-01138-2

Med Phys. 2024 Jun 6. doi: 10.1002/mp.17232. Online ahead of print.

ABSTRACT

BACKGROUND: Computer algorithms that simulate lower-doses computed tomography (CT) images from clinical-dose images are widely available. However, most operate in the projection domain and assume access to the reconstruction method. Access to commercial reconstruction methods may often not be available in medical research, making image-domain noise simulation methods useful. However, the introduction of non-linear reconstruction methods, such as iterative and deep learning-based reconstruction, makes noise insertion in the image domain intractable, as it is not possible to determine the noise textures analytically.

PURPOSE: To develop a deep learning-based image-domain method to generate low-dose CT images from clinical-dose CT (CDCT) images for non-linear reconstruction methods.

METHODS: We propose a fully image domain-based method, utilizing a series of three convolutional neural networks (CNNs), which, respectively, denoise CDCT images, predict the standard deviation map of the low-dose image, and generate the noise power spectra (NPS) of local patches throughout the low-dose image. All three models have U-net-based architectures and are partly or fully three-dimensional. As a use case for this study and with no loss of generality, we use paired low-dose and clinical-dose brain CT scans. A dataset of 326 $\hskip.001pt 326$ paired scans was retrospectively obtained. All images were acquired with a wide-area detector clinical system and reconstructed using its standard clinical iterative algorithm. Each pair was registered using rigid registration to correct for motion between acquisitions. The data was randomly partitioned into training ( 251 $\hskip.001pt 251$ samples), validation ( 25 $\hskip.001pt 25$ samples), and test ( 50 $\hskip.001pt 50$ samples) sets. The performance of each of these three CNNs was validated separately. For the denoising CNN, the local standard deviation decrease, and bias were determined. For the standard deviation map CNN, the real and estimated standard deviations were compared locally. Finally, for the NPS CNN, the NPS of the synthetic and real low-dose noise were compared inside and outside the skull. Two proof-of-concept denoising studies were performed to determine if the performance of a CNN- or a gradient-based denoising filter on the synthetic low-dose data versus real data differed.

RESULTS: The denoising network had a median decrease in noise in the cerebrospinal fluid by a factor of 1.71 $1.71$ and introduced a median bias of + 0.7 $ + 0.7$ HU. The network for standard deviation map estimation had a median error of + 0.1 $ + 0.1$ HU. The noise power spectrum estimation network was able to capture the anisotropic and shift-variant nature of the noise structure by showing good agreement between the synthetic and real low-dose noise and their corresponding power spectra. The two proof of concept denoising studies showed only minimal difference in standard deviation improvement ratio between the synthetic and real low-dose CT images with the median difference between the two being 0.0 and +0.05 for the CNN- and gradient-based filter, respectively.

CONCLUSION: The proposed method demonstrated good performance in generating synthetic low-dose brain CT scans without access to the projection data or to the reconstruction method. This method can generate multiple low-dose image realizations from one clinical-dose image, so it is useful for validation, optimization, and repeatability studies of image-processing algorithms.

PMID:38843540 | DOI:10.1002/mp.17232

J Clin Oncol. 2024 Jun 6:JCO2301978. doi: 10.1200/JCO.23.01978. Online ahead of print.

ABSTRACT

PURPOSE: Artificial intelligence can reduce the time used by physicians on radiological assessments. For 18F-fluorodeoxyglucose-avid lymphomas, obtaining complete metabolic response (CMR) by end of treatment is prognostic.

METHODS: Here, we present a deep learning-based algorithm for fully automated treatment response assessments according to the Lugano 2014 classification. The proposed four-stage method, trained on a multicountry clinical trial (ClinicalTrials.gov identifier: NCT01287741) and tested in three independent multicenter and multicountry test sets on different non-Hodgkin lymphoma subtypes and different lines of treatment (ClinicalTrials.gov identifiers: NCT02257567, NCT02500407, 20% holdout in ClinicalTrials.gov identifier: NCT01287741), outputs the detected lesions at baseline and follow-up to enable focused radiologist review.

RESULTS: The method's response assessment achieved high agreement with the adjudicated radiologic responses (eg, agreement for overall response assessment of 93%, 87%, and 85% in ClinicalTrials.gov identifiers: NCT01287741, NCT02500407, and NCT02257567, respectively) similar to inter-radiologist agreement and was strongly prognostic of outcomes with a trend toward higher accuracy for death risk than adjudicated radiologic responses (hazard ratio for end of treatment by-model CMR of 0.123, 0.054, and 0.205 in ClinicalTrials.gov identifiers: NCT01287741, NCT02500407, and NCT02257567, compared with, respectively, 0.226, 0.292, and 0.272 for CMR by the adjudicated responses). Furthermore, a radiologist review of the algorithm's assessments was conducted. The radiologist median review time was 1.38 minutes/assessment, and no statistically significant differences were observed in the level of agreement of the radiologist with the model's response compared with the level of agreement of the radiologist with the adjudicated responses.

CONCLUSION: These results suggest that the proposed method can be incorporated into radiologic response assessment workflows in cancer imaging for significant time savings and with performance similar to trained medical experts.

PMID:38843483 | DOI:10.1200/JCO.23.01978

ACS Sens. 2024 Jun 6. doi: 10.1021/acssensors.4c00488. Online ahead of print.

ABSTRACT

An integrated approach combining surface-enhanced Raman spectroscopy (SERS) with a specialized deep learning algorithm to rapidly and accurately detect and quantify SARS-CoV-2 variants is developed based on an angiotensin-converting enzyme 2 (ACE2)-functionalized AgNR@SiO2 array SERS sensor. SERS spectra with concentrations of different variants were collected using a portable Raman system. After appropriate spectral preprocessing, a deep learning algorithm, CoVari, is developed to predict both the viral variant species and concentrations. Using a 10-fold cross-validation strategy, the model achieves an average accuracy of 99.9% in discriminating between different virus variants and R2 values larger than 0.98 for quantifying viral concentrations of the three viruses, demonstrating the high quality of the detection. The limit of detection of the ACE2 SERS sensor is determined to be 10.472, 11.882, and 21.591 PFU/mL for SARS-CoV-2, SARS-CoV-2 B1, and CoV-NL63, respectively. The feature importance of virus classification and concentration regression in the CoVari algorithm are calculated based on a permutation algorithm, which showed a clear correlation to the biochemical origins of the spectra or spectral changes. In an unknown specimen test, classification accuracy can achieve >90% for concentrations larger than 781 PFU/mL, and the predicted concentrations consistently align with actual values, highlighting the robustness of the proposed algorithm. Based on the CoVari architecture and the output vector, this algorithm can be generalized to predict both viral variant species and concentrations simultaneously for a broader range of viruses. These results demonstrate that the SERS + CoVari strategy has the potential for rapid and quantitative detection of virus variants and potentially point-of-care diagnostic platforms.

PMID:38843447 | DOI:10.1021/acssensors.4c00488

PLoS One. 2024 Jun 6;19(6):e0290915. doi: 10.1371/journal.pone.0290915. eCollection 2024.

ABSTRACT

The Urdu language is spoken and written on different social media platforms like Twitter, WhatsApp, Facebook, and YouTube. However, due to the lack of Urdu Language Processing (ULP) libraries, it is quite challenging to identify threats from textual and sequential data on the social media provided in Urdu. Therefore, it is required to preprocess the Urdu data as efficiently as English by creating different stemming and data cleaning libraries for Urdu data. Different lexical and machine learning-based techniques are introduced in the literature, but all of these are limited to the unavailability of online Urdu vocabulary. This research has introduced Urdu language vocabulary, including a stop words list and a stemming dictionary to preprocess Urdu data as efficiently as English. This reduced the input size of the Urdu language sentences and removed redundant and noisy information. Finally, a deep sequential model based on Long Short-Term Memory (LSTM) units is trained on the efficiently preprocessed, evaluated, and tested. Our proposed methodology resulted in good prediction performance, i.e., an accuracy of 82%, which is greater than the existing methods.

PMID:38843283 | DOI:10.1371/journal.pone.0290915

PLoS One. 2024 Jun 6;19(6):e0303890. doi: 10.1371/journal.pone.0303890. eCollection 2024.

ABSTRACT

Anomaly detection in time series data is essential for fraud detection and intrusion monitoring applications. However, it poses challenges due to data complexity and high dimensionality. Industrial applications struggle to process high-dimensional, complex data streams in real time despite existing solutions. This study introduces deep ensemble models to improve traditional time series analysis and anomaly detection methods. Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks effectively handle variable-length sequences and capture long-term relationships. Convolutional Neural Networks (CNNs) are also investigated, especially for univariate or multivariate time series forecasting. The Transformer, an architecture based on Artificial Neural Networks (ANN), has demonstrated promising results in various applications, including time series prediction and anomaly detection. Graph Neural Networks (GNNs) identify time series anomalies by capturing temporal connections and interdependencies between periods, leveraging the underlying graph structure of time series data. A novel feature selection approach is proposed to address challenges posed by high-dimensional data, improving anomaly detection by selecting different or more critical features from the data. This approach outperforms previous techniques in several aspects. Overall, this research introduces state-of-the-art algorithms for anomaly detection in time series data, offering advancements in real-time processing and decision-making across various industrial sectors.

PMID:38843255 | DOI:10.1371/journal.pone.0303890

IEEE J Biomed Health Inform. 2024 Jun;28(6):3379-3388. doi: 10.1109/JBHI.2024.3384453.

ABSTRACT

Monitoring in-bed pose estimation based on the Internet of Medical Things (IoMT) and ambient technology has a significant impact on many applications such as sleep-related disorders including obstructive sleep apnea syndrome, assessment of sleep quality, and health risk of pressure ulcers. In this research, a new multimodal in-bed pose estimation has been proposed using a deep learning framework. The Simultaneously-collected multimodal Lying Pose (SLP) dataset has been used for performance evaluation of the proposed framework where two modalities including long wave infrared (LWIR) and depth images are used to train the proposed model. The main contribution of this research is the feature fusion network and the use of a generative model to generate RGB images having similar poses to other modalities (LWIR/depth). The inclusion of a generative model helps to improve the overall accuracy of the pose estimation algorithm. Moreover, the method can be generalized for situations to recover human pose both in home and hospital settings under various cover thickness levels. The proposed model is compared with other fusion-based models and shows an improved performance of 97.8% at PCKh @0.5. In addition, performance has been evaluated for different cover conditions, and under home and hospital environments which present improvements using our proposed model.

PMID:38843069 | DOI:10.1109/JBHI.2024.3384453

IEEE Trans Neural Netw Learn Syst. 2024 Jun 6;PP. doi: 10.1109/TNNLS.2024.3400928. Online ahead of print.

ABSTRACT

Low-rankness plays an important role in traditional machine learning but is not so popular in deep learning. Most previous low-rank network compression methods compress networks by approximating pretrained models and retraining. However, the optimal solution in the Euclidean space may be quite different from the one with low-rank constraint. A well-pretrained model is not a good initialization for the model with low-rank constraints. Thus, the performance of a low-rank compressed network degrades significantly. Compared with other network compression methods such as pruning, low-rank methods attract less attention in recent years. In this article, we devise a new training method, low-rank projection with energy transfer (LRPET), that trains low-rank compressed networks from scratch and achieves competitive performance. We propose to alternately perform stochastic gradient descent training and projection of each weight matrix onto the corresponding low-rank manifold. Compared to retraining on the compact model, this enables full utilization of model capacity since solution space is relaxed back to Euclidean space after projection. The matrix energy (the sum of squares of singular values) reduction caused by projection is compensated by energy transfer. We uniformly transfer the energy of the pruned singular values to the remaining ones. We theoretically show that energy transfer eases the trend of gradient vanishing caused by projection. In modern networks, a batch normalization (BN) layer can be merged into the previous convolution layer for inference, thereby influencing the optimal low-rank approximation (LRA) of the previous layer. We propose BN rectification to cut off its effect on the optimal LRA, which further improves the performance. Comprehensive experiments on CIFAR-10 and ImageNet have justified that our method is superior to other low-rank compression methods and also outperforms recent state-of-the-art pruning methods. For object detection and semantic segmentation, our method still achieves good compression results. In addition, we combine LRPET with quantization and hashing methods and achieve even better compression than the original single method. We further apply it in Transformer-based models to demonstrate its transferability. Our code is available at https://github.com/BZQLin/LRPET.

PMID:38843062 | DOI:10.1109/TNNLS.2024.3400928

IEEE Trans Ultrason Ferroelectr Freq Control. 2024 Jun 6;PP. doi: 10.1109/TUFFC.2024.3410671. Online ahead of print.

ABSTRACT

Ultrasound elastography is a non-invasive medical imaging technique that maps viscoelastic properties to characterize tissues and diseases. Elastography can be divided into two classes in a broad sense: strain elastography (SE), which relies on Hooke's law to delineate strain as a surrogate for elasticity, and shear-wave elastography (SWE), which tracks the propagation of shear waves in tissues to estimate the elasticity. As tracking the displacement field in the temporal or spatial domain is an inevitable step of both SE and SWE, the success is contingent on the displacement estimation accuracy. Recent reviews mostly focused on clinical applications of elastography, disregarding advances in displacement tracking algorithms. Herein, we comprehensively review the recently proposed displacement estimation algorithms applied to both SE and SWE. In addition to cross-correlation, block-matching (i.e., window-based), model-based, energy-based, and deep learning-based tracking techniques, we review large and lateral displacement tracking, adaptive beamforming, data enhancement, and noise-suppression algorithms facilitating better displacement estimation. We also discuss the simulation models for displacement tracking validation, clinical translation and validation of displacement tracking methods, performance evaluation metrics, and publicly available codes and data for displacement tracking in elastography. Finally, we provide experiential opinions on different tracking algorithms, list the limitations of the current state of elastographic tracking, and comment on possible future research.

PMID:38843058 | DOI:10.1109/TUFFC.2024.3410671

Niger J Clin Pract. 2024 May 1;27(5):669-677. doi: 10.4103/njcp.njcp_220_24. Epub 2024 May 29.

ABSTRACT

BACKGROUND: Panoramic radiography (PR) is available to determine the contact relationship between maxillary molar teeth (MMT) and the maxillary sinus floor (MSF). However, as PRs do not provide clear and detailed anatomical information, advanced imaging methods can be used.

AIM: The aim of this study was to evaluate the diagnostic performance of deep learning (DL) applications that assess the relationship of the MSF to the first maxillary molar teeth (fMMT) and second maxillary molar teeth (sMMT) on PRs with data confirmed by cone beam computed tomography (CBCT).

METHODS: A total of 2162 fMMT and sMMT were included in this retrospective study. The contact relationship of teeth with MSF was compared among DL methods.

RESULTS: DL methods, such as GoogLeNet, VGG16, VGG19, DarkNet19, and DarkNet53, were used to evaluate the contact relationship between MMT and MSF, and 85.89% accuracy was achieved by majority voting. In addition, 88.72%, 81.19%, 89.39%, and 83.14% accuracy rates were obtained in right fMMT, right sMMT, left fMMT, and left sMMT, respectively.

CONCLUSION: DL models showed high accuracy values in detecting the relationship of fMMT and sMMT with MSF.

PMID:38842718 | DOI:10.4103/njcp.njcp_220_24

Eur Radiol. 2024 Jun 6. doi: 10.1007/s00330-024-10812-6. Online ahead of print.

ABSTRACT

OBJECTIVES: To develop an automated pipeline for extracting prostate cancer-related information from clinical notes.

MATERIALS AND METHODS: This retrospective study included 23,225 patients who underwent prostate MRI between 2017 and 2022. Cancer risk factors (family history of cancer and digital rectal exam findings), pre-MRI prostate pathology, and treatment history of prostate cancer were extracted from free-text clinical notes in English as binary or multi-class classification tasks. Any sentence containing pre-defined keywords was extracted from clinical notes within one year before the MRI. After manually creating sentence-level datasets with ground truth, Bidirectional Encoder Representations from Transformers (BERT)-based sentence-level models were fine-tuned using the extracted sentence as input and the category as output. The patient-level output was determined by compilation of multiple sentence-level outputs using tree-based models. Sentence-level classification performance was evaluated using the area under the receiver operating characteristic curve (AUC) on 15% of the sentence-level dataset (sentence-level test set). The patient-level classification performance was evaluated on the patient-level test set created by radiologists by reviewing the clinical notes of 603 patients. Accuracy and sensitivity were compared between the pipeline and radiologists.

RESULTS: Sentence-level AUCs were ≥ 0.94. The pipeline showed higher patient-level sensitivity for extracting cancer risk factors (e.g., family history of prostate cancer, 96.5% vs. 77.9%, p < 0.001), but lower accuracy in classifying pre-MRI prostate pathology (92.5% vs. 95.9%, p = 0.002) and treatment history of prostate cancer (95.5% vs. 97.7%, p = 0.03) than radiologists, respectively.

CONCLUSION: The proposed pipeline showed promising performance, especially for extracting cancer risk factors from patient's clinical notes.

CLINICAL RELEVANCE STATEMENT: The natural language processing pipeline showed a higher sensitivity for extracting prostate cancer risk factors than radiologists and may help efficiently gather relevant text information when interpreting prostate MRI.

KEY POINTS: When interpreting prostate MRI, it is necessary to extract prostate cancer-related information from clinical notes. This pipeline extracted the presence of prostate cancer risk factors with higher sensitivity than radiologists. Natural language processing may help radiologists efficiently gather relevant prostate cancer-related text information.

PMID:38842692 | DOI:10.1007/s00330-024-10812-6

Ann Nucl Med. 2024 Jun 6. doi: 10.1007/s12149-024-01945-1. Online ahead of print.

ABSTRACT

BACKGROUND: Cardiac positron emission tomography (PET) can visualize and quantify the molecular and physiological pathways of cardiac function. However, cardiac and respiratory motion can introduce blurring that reduces PET image quality and quantitative accuracy. Dual cardiac- and respiratory-gated PET reconstruction can mitigate motion artifacts but increases noise as only a subset of data are used for each time frame of the cardiac cycle.

AIM: The objective of this study is to create a zero-shot image denoising framework using a conditional generative adversarial networks (cGANs) for improving image quality and quantitative accuracy in non-gated and dual-gated cardiac PET images.

METHODS: Our study included retrospective list-mode data from 40 patients who underwent an 18F-fluorodeoxyglucose (18F-FDG) cardiac PET study. We initially trained and evaluated a 3D cGAN-known as Pix2Pix-on simulated non-gated low-count PET data paired with corresponding full-count target data, and then deployed the model on an unseen test set acquired on the same PET/CT system including both non-gated and dual-gated PET data.

RESULTS: Quantitative analysis demonstrated that the 3D Pix2Pix network architecture achieved significantly (p value<0.05) enhanced image quality and accuracy in both non-gated and gated cardiac PET images. At 5%, 10%, and 15% preserved count statistics, the model increased peak signal-to-noise ratio (PSNR) by 33.7%, 21.2%, and 15.5%, structural similarity index (SSIM) by 7.1%, 3.3%, and 2.2%, and reduced mean absolute error (MAE) by 61.4%, 54.3%, and 49.7%, respectively. When tested on dual-gated PET data, the model consistently reduced noise, irrespective of cardiac/respiratory motion phases, while maintaining image resolution and accuracy. Significant improvements were observed across all gates, including a 34.7% increase in PSNR, a 7.8% improvement in SSIM, and a 60.3% reduction in MAE.

CONCLUSION: The findings of this study indicate that dual-gated cardiac PET images, which often have post-reconstruction artifacts potentially affecting diagnostic performance, can be effectively improved using a generative pre-trained denoising network.

PMID:38842629 | DOI:10.1007/s12149-024-01945-1

Brief Bioinform. 2024 May 23;25(4):bbae270. doi: 10.1093/bib/bbae270.

ABSTRACT

Peptide- and protein-based therapeutics are becoming a promising treatment regimen for myriad diseases. Toxicity of proteins is the primary hurdle for protein-based therapies. Thus, there is an urgent need for accurate in silico methods for determining toxic proteins to filter the pool of potential candidates. At the same time, it is imperative to precisely identify non-toxic proteins to expand the possibilities for protein-based biologics. To address this challenge, we proposed an ensemble framework, called VISH-Pred, comprising models built by fine-tuning ESM2 transformer models on a large, experimentally validated, curated dataset of protein and peptide toxicities. The primary steps in the VISH-Pred framework are to efficiently estimate protein toxicities taking just the protein sequence as input, employing an under sampling technique to handle the humongous class-imbalance in the data and learning representations from fine-tuned ESM2 protein language models which are then fed to machine learning techniques such as Lightgbm and XGBoost. The VISH-Pred framework is able to correctly identify both peptides/proteins with potential toxicity and non-toxic proteins, achieving a Matthews correlation coefficient of 0.737, 0.716 and 0.322 and F1-score of 0.759, 0.696 and 0.713 on three non-redundant blind tests, respectively, outperforming other methods by over $10\%$ on these quality metrics. Moreover, VISH-Pred achieved the best accuracy and area under receiver operating curve scores on these independent test sets, highlighting the robustness and generalization capability of the framework. By making VISH-Pred available as an easy-to-use web server, we expect it to serve as a valuable asset for future endeavors aimed at discerning the toxicity of peptides and enabling efficient protein-based therapeutics.

PMID:38842509 | DOI:10.1093/bib/bbae270

J Chem Theory Comput. 2024 Jun 6. doi: 10.1021/acs.jctc.4c00422. Online ahead of print.

ABSTRACT

Computer prediction of NMR chemical shifts plays an increasingly important role in molecular structure assignment and elucidation for organic molecule studies. Density functional theory (DFT) and gauge-including atomic orbital (GIAO) have established a framework to predict NMR chemical shifts but often at a significant computational expense with a limited prediction accuracy. Recent advancements in deep learning methods, especially graph neural networks (GNNs), have shown promise in improving the accuracy of predicting experimental chemical shifts, either by using 2D molecular topological features or 3D conformational representation. This study presents a new 3D GNN model to predict 1H and 13C chemical shifts, CSTShift, that combines atomic features with DFT-calculated shielding tensor descriptors, capturing both isotropic and anisotropic shielding effects. Utilizing the NMRShiftDB2 data set and conducting DFT optimization and GIAO calculations at the B3LYP/6-31G(d) level, we prepared the NMRShiftDB2-DFT data set of high-quality 3D structures and shielding tensors with corresponding experimentally measured 1H and 13C chemical shifts. The developed CSTShift models achieve the state-of-the-art prediction performance on both the NMRShiftDB2-DFT test data set and external CHESHIRE data set. Further case studies on identifying correct structures from two groups of constitutional isomers show its capability for structure assignment and elucidation. The source code and data are accessible at https://yzhang.hpc.nyu.edu/IMA.

PMID:38842505 | DOI:10.1021/acs.jctc.4c00422

Radiol Cardiothorac Imaging. 2024 Jun;6(3):e240140. doi: 10.1148/ryct.240140.

NO ABSTRACT

PMID:38842457 | DOI:10.1148/ryct.240140

ACS Sens. 2024 Jun 6. doi: 10.1021/acssensors.4c00654. Online ahead of print.

ABSTRACT

The identification of drug-induced cardiotoxicity remains a pressing challenge with far-reaching clinical and economic ramifications, often leading to patient harm and resource-intensive drug recalls. Current methodologies, including in vivo and in vitro models, have severe limitations in accurate identification of cardiotoxic substances. Pioneering a paradigm shift from these conventional techniques, our study presents two deep learning-based frameworks, STFT-CNN and SST-CNN, to assess cardiotoxicity with markedly improved accuracy and reliability. Leveraging the power of induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) as a more human-relevant cell model, we record mechanical beating signals through impedance measurements. These temporal signals were converted into enriched two-dimensional representations through advanced transformation techniques, specifically short-time Fourier transform (STFT) and synchro-squeezing transform (SST). These transformed data are fed into the proposed frameworks for comprehensive analysis, including drug type classification, concentration classification, cardiotoxicity classification, and new drug identification. Compared to traditional models like recurrent neural network (RNN) and 1-dimensional convolutional neural network (1D-CNN), SST-CNN delivered an impressive test accuracy of 98.55% in drug type classification and 99% in distinguishing cardiotoxic and noncardiotoxic drugs. Its feasibility is further highlighted with a stellar 98.5% average accuracy for classification of various concentrations, and the superiority of our proposed frameworks underscores their promise in revolutionizing drug safety assessments. With a potential for scalability, they represent a major leap in drug safety assessments, offering a pathway to more robust, efficient, and human-relevant cardiotoxicity evaluations.

PMID:38842187 | DOI:10.1021/acssensors.4c00654