Front Neurosci. 2024 Sep 30;18:1416174. doi: 10.3389/fnins.2024.1416174. eCollection 2024.

ABSTRACT

BACKGROUND: White matter hyperintensities (WMH) observed in T2 fluid-attenuated inversion recovery (FLAIR) images have emerged as potential markers of neurodegenerative diseases like Multiple Sclerosis (MS). Lacking comprehensive automated WMH classification systems in current research, there is a need to develop accurate detection and classification methods for WMH that will benefit the diagnosis and monitoring of brain diseases.

OBJECTIVE: Juxtacortical WMH (JCWMH) is a less explored subtype of WMH, primarily due to the hard definition of the cortex in FLAIR images, which is escalated by the presence of lesions to obtain appropriate gray matter (GM) masks.

METHODS: In this study, we present a method to perform a specialized GM segmentation developed for the classification of WMH, especially JCWMH. Using T1 and FLAIR images, we propose a pipeline to integrate masks of white matter, cerebrospinal fluid, ventricles, and WMH to create a unique mask to refine the primary GM map. Subsequently, we utilize this pipeline to generate paired data for training a conditional generative adversarial network (cGAN) to substitute the pipeline and reduce the inputs to only FLAIR images. The classification of WMH is then based on the distances between WMH and ventricular and GM masks. Due to the lack of multi-class labeled WMH datasets and the need for extensive data for training deep learning models, we attempted to collect a large local dataset and manually segment and label some data for WMH and ventricles.

RESULTS: In JCWMH classification, the proposed method exhibited a Dice similarity coefficient, precision, and sensitivity of 0.76, 0.69, and 0.84, respectively. With values of 0.66, 0.55, and 0.81, the proposed method clearly outperformed the approach commonly used in the literature, which uses extracted GM masks from registered T1 images on FLAIR.

CONCLUSION: After training, the method proves its efficiency by providing results in less than one second. In contrast, the usual approach would require at least two minutes for registration and segmentation alone. The proposed method is automated and fast and requires no initialization as it works exclusively with FLAIR images. Such innovative methods will undoubtedly facilitate accurate and meaningful analysis of WMH in clinical practice by reducing complexity and increasing efficiency.

PMID:39403067 | PMC:PMC11471731 | DOI:10.3389/fnins.2024.1416174

J Am Soc Mass Spectrom. 2024 Oct 14. doi: 10.1021/jasms.4c00186. Online ahead of print.

ABSTRACT

The development of matrices for Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry (MALDI MS) has traditionally relied on experimental efforts. Here, we propose a Goal-Directed artificial intelligence generative model, fueled by computational chemistry calculated data, to construct a chemical space optimized for Electron Transfer (ET) processes in MALDI analysis. We utilized a group of 30 reported ET matrices, subjected to structural enumeration and molecular properties prediction using semiempirical and ab initio calculations, to establish a comprehensive database comprising diverse structural and property data. Subsequently, employing a protocol of structural enumeration with 68 canonical SMILES of Bemis-Murcko (BM) fragments, we expanded the structural complexity of the initial library. This process generated 82753 compounds organized into 10 scaffold levels, with a p50 index from the Cyclic System Retrieval (CSR) curve of scaffolds of 50%. From the resulting enumerated library, a diverse subset of structures was selected by using the Jarvis-Patrick clustering method. These structures, along with their associated properties measured from quantum mechanics and experimental data, were used to train a Machine Learning (ML) model to predict ionization energy (Ei) values. Subsequently, a Scoring Neural Network (SNN), coupled with our Goal-Directed generative model using a Recurrent Neural Network (RNN) with Deep Learning (DL) architectures, was trained. The generative model was guided using a prior network within a Reinforcement/Transfer Learning environment. The final AI-generative model learned that structures with high unsaturation, H/C ratios under 1, and molecular weights between 100 and 300 u are favorable for ET MALDI matrices, as well as those with few aromatic rings and zero aliphatic rings. Other molecular features were also favored. The resulting AI-generated library exhibits Ei values over 8.0 eV, akin to those of reported "good" ET MALDI matrices, indicating successful design with high synthesis accessibility scores. In conclusion, our generative model provided valuable insights into the molecular features ideal for ET MALDI compounds while generating a wide range of structurally diverse molecules within a similar molecular property space. The next critical step in this process is to synthesize a selection of these generated compounds for the experimental validation and further characterization.

PMID:39402868 | DOI:10.1021/jasms.4c00186

Clin Transl Sci. 2024 Oct;17(10):e70045. doi: 10.1111/cts.70045.

ABSTRACT

Population pharmacokinetic (PK) modeling serves as the cornerstone for understanding drug behavior within a specific population. It integrates subject covariates to elucidate the variability in PK parameters, thus enhancing predictive accuracy. However, covariate modeling within this framework can be intricate and time-consuming due to the often obscure structural relationship between covariates and PK parameters. Previous attempts, such as deep compartment modeling (DCM), aimed to streamline this process using machine learning techniques. Nonetheless, DCM fell short in assessing residual errors and interindividual variability (IIV), potentially leading to model misspecification and overfitting. Furthermore, DCM lacked the ability to quantify model uncertainty. To address these limitations, we introduce hierarchical deep compartment modeling (HDCM) as an advancement of DCM. HDCM harnesses machine learning to discern the interplay between covariates and PK parameters while simultaneously evaluating diverse levels of random effects and quantifying uncertainty through Bayesian inference. This tutorial provides a comprehensive application of the HDCM workflow using open-source Julia tools.

PMID:39402751 | DOI:10.1111/cts.70045

BMC Res Notes. 2024 Oct 14;17(1):306. doi: 10.1186/s13104-024-06959-6.

ABSTRACT

OBJECTIVES: Recent advances in bio-inspired navigation have sparked interest in the phenomenon of skylight polarization. This interest stems from the potential of skylight-based orientation sensors, which performance can be simulated using physical models. However, the effectiveness of machine learning algorithms in this domain relies heavily on access to large datasets for training. Although there are several databases of simulated images in literature, there remains a lack of publicly available annotated real-world color polarimetric images of the sky across various weather conditions.

DATA DESCRIPTION: We present here a dataset obtained from a long-term experimental setup designed to collect polarimetric images from a stand-alone camera. The setup utilizes a Division-of-Focal-Plane polarization camera equipped with a fisheye lens mounted on a rotative telescope mount. Furthermore, we obtained the sensor's orientation within the East-North-Up (ENU) frame from a geometrical calibration and an algorithm provided with the database. To facilitate further research in this area, the present sample dataset spanning two months has been made available on a public archive with manual annotations as required by deep learning algorithms. The images were acquired at 10 min intervals and were taken with various exposure times ranging from 33µs to 300ms.

PMID:39402604 | DOI:10.1186/s13104-024-06959-6

BMC Genomics. 2024 Oct 14;25(1):957. doi: 10.1186/s12864-024-10859-1.

ABSTRACT

BACKGROUND: Transcription factors (TFs) bind to different parts of the genome in different types of cells, but it is usually assumed that the inherent DNA-binding preferences of a TF are invariant to cell type. Yet, there are several known examples of TFs that switch their DNA-binding preferences in different cell types, and yet more examples of other mechanisms, such as steric hindrance or cooperative binding, that may result in a "DNA signature" of differential binding.

RESULTS: To survey this phenomenon systematically, we developed a deep learning method we call SigTFB (Signatures of TF Binding) to detect and quantify cell-type specificity in a TF's known genomic binding sites. We used ENCODE ChIP-seq data to conduct a wide scale investigation of 169 distinct TFs in up to 14 distinct cell types. SigTFB detected statistically significant DNA binding signatures in approximately two-thirds of TFs, far more than might have been expected from the relatively sparse evidence in prior literature. We found that the presence or absence of a cell-type specific DNA binding signature is distinct from, and indeed largely uncorrelated to, the degree of overlap between ChIP-seq peaks in different cell types, and tended to arise by two mechanisms: using established motifs in different frequencies, and by selective inclusion of motifs for distint TFs.

CONCLUSIONS: While recent results have highlighted cell state features such as chromatin accessibility and gene expression in predicting TF binding, our results emphasize that, for some TFs, the DNA sequences of the binding sites contain substantial cell-type specific motifs.

PMID:39402535 | DOI:10.1186/s12864-024-10859-1

BMC Pulm Med. 2024 Oct 14;24(1):515. doi: 10.1186/s12890-024-03333-x.

ABSTRACT

BACKGROUND: Previous studies have shown that patients with pre-existing chronic obstructive pulmonary diseases (COPD) were more likely to be infected with coronavirus disease (COVID-19) and lead to more severe lung lesions. However, few studies have explored the severity and prognosis of COVID-19 patients with different phenotypes of COPD.

PURPOSE: The aim of this study is to investigate the value of the deep learning and radiomics features for the severity evaluation and the nucleic acid turning-negative time prediction in COVID-19 patients with COPD including two phenotypes of chronic bronchitis predominant patients and emphysema predominant patients.

METHODS: A total of 281 patients were retrospectively collected from Hohhot First Hospital between October 2022 and January 2023. They were divided to three groups: COVID-19 group of 95 patients, COVID-19 with emphysema group of 94 patients, COVID-19 with chronic bronchitis group of 92 patients. All patients underwent chest computed tomography (CT) scans and recorded clinical data. The U-net model was pretrained to segment the pulmonary involvement area on CT images and the severity of pneumonia were evaluated by the percentage of pulmonary involvement volume to lung volume. The 107 radiomics features were extracted by pyradiomics package. The Spearman method was employed to analyze the correlation of the data and visualize it through a heatmap. Then we establish a deep learning model (model 1) and a fusion model (model 2) combined deep learning with radiomics features to predict nucleic acid turning-negative time.

RESULTS: COVID-19 patients with emphysema was lowest in the lymphocyte count compared to COVID-19 patients and COVID-19 companied with chronic bronchitis, and they have the most extensive range of pulmonary inflammation. The lymphocyte count was significantly correlated with pulmonary involvement and the time for nucleic acid turning negative (r=-0.145, P < 0.05). Importantly, our results demonstrated that model 2 achieved an accuracy of 80.9% in predicting nucleic acid turning-negative time.

CONCLUSION: The pre-existing emphysema phenotype of COPD severely aggravated the pulmonary involvement of COVID-19 patients. Deep learning and radiomics features may provide more information to accurately predict the nucleic acid turning-negative time, which is expected to play an important role in clinical practice.

PMID:39402509 | DOI:10.1186/s12890-024-03333-x

BMC Bioinformatics. 2024 Oct 14;25(1):328. doi: 10.1186/s12859-024-05955-8.

ABSTRACT

BACKGROUND: The rapid advancements in deep neural network models have significantly enhanced the ability to extract features from microbial sequence data, which is critical for addressing biological challenges. However, the scarcity and complexity of labeled microbial data pose substantial difficulties for supervised learning approaches. To address these issues, we propose DNASimCLR, an unsupervised framework designed for efficient gene sequence data feature extraction.

RESULTS: DNASimCLR leverages convolutional neural networks and the SimCLR framework, based on contrastive learning, to extract intricate features from diverse microbial gene sequences. Pre-training was conducted on two classic large scale unlabelled datasets encompassing metagenomes and viral gene sequences. Subsequent classification tasks were performed by fine-tuning the pretrained model using the previously acquired model. Our experiments demonstrate that DNASimCLR is at least comparable to state-of-the-art techniques for gene sequence classification. For convolutional neural network-based approaches, DNASimCLR surpasses the latest existing methods, clearly establishing its superiority over the state-of-the-art CNN-based feature extraction techniques. Furthermore, the model exhibits superior performance across diverse tasks in analyzing biological sequence data, showcasing its robust adaptability.

CONCLUSIONS: DNASimCLR represents a robust and database-agnostic solution for gene sequence classification. Its versatility allows it to perform well in scenarios involving novel or previously unseen gene sequences, making it a valuable tool for diverse applications in genomics.

PMID:39402441 | DOI:10.1186/s12859-024-05955-8

J Imaging Inform Med. 2024 Oct 14. doi: 10.1007/s10278-024-01297-2. Online ahead of print.

ABSTRACT

Breast cancer is the most common cancer in women globally, imposing a significant burden on global public health due to high death rates. Data from the World Health Organization show an alarming annual incidence of nearly 2.3 million new cases, drawing the attention of patients, healthcare professionals, and governments alike. Through the examination of histopathological pictures, this study aims to revolutionize the early and precise identification of breast cancer by utilizing the capabilities of a deep convolutional neural network (CNN)-based model. The model's performance is improved by including numerous classifiers, including support vector machine (SVM), decision tree, and K-nearest neighbors (KNN), using transfer learning techniques. The studies include evaluating two separate feature vectors, one with and one without principal component analysis (PCA). Extensive comparisons are made to measure the model's performance against current deep learning models, including critical metrics such as false positive rate, true positive rate, accuracy, precision, and recall. The data show that the SVM algorithm with PCA features achieves excellent speed and accuracy, with an amazing accuracy of 99.5%. Furthermore, although being somewhat slower than SVM, the decision tree model has the greatest accuracy of 99.4% without PCA. This study suggests a viable strategy for improving early breast cancer diagnosis, opening the path for more effective healthcare treatments and better patient outcomes.

PMID:39402357 | DOI:10.1007/s10278-024-01297-2

J Imaging Inform Med. 2024 Oct 14. doi: 10.1007/s10278-024-01256-x. Online ahead of print.

ABSTRACT

Radiation dose and image quality in radiology are influenced by the X-ray prime factors: KVp, mAs, and source-detector distance. These parameters are set by the X-ray technician prior to the acquisition considering the radiographic position. A wrong setting of these parameters may result in exposure errors, forcing the test to be repeated with the increase of the radiation dose delivered to the patient. This work presents a novel approach based on deep learning that automatically estimates the radiographic position from a photograph captured prior to X-ray exposure, which can then be used to select the optimal prime factors. We created a database using 66 radiographic positions commonly used in clinical settings, prospectively obtained during 2022 from 75 volunteers in two different X-ray facilities. The architecture for radiographic position classification was a lightweight version of ConvNeXt trained with fine-tuning, discriminative learning rates, and a one-cycle policy scheduler. Our resulting model achieved an accuracy of 93.17% for radiographic position classification and increased to 95.58% when considering the correct selection of prime factors, since half of the errors involved positions with the same KVp and mAs values. Most errors occurred for radiographic positions with similar patient pose in the photograph. Results suggest the feasibility of the method to facilitate the acquisition workflow reducing the occurrence of exposure errors while preventing unnecessary radiation dose delivered to patients.

PMID:39402356 | DOI:10.1007/s10278-024-01256-x

Urolithiasis. 2024 Oct 14;52(1):145. doi: 10.1007/s00240-024-01644-6.

ABSTRACT

Kidney stones are the most common urinary system diseases, and early identification is of great significance. The purpose of this study was to use routine urine and blood detection indices to build a deep learning (DL) model to identify the presence of kidney stones in the early stage. A retrospective analysis was conducted on patients with kidney stones who were treated at West China Hospital of Sichuan University from January 2020 to June 2023. A total of 1130 individuals presenting with kidney stones and 1230 healthy subjects were enrolled. The first blood and urine laboratory data of participants at our hospital were collected, and the data were divided into a training dataset (80%) and a verification dataset (20%). Additionally, a long short-term memory (LSTM)-based adaptive feature weighting model was trained for the early identification of kidney stones, and the results were compared with those of other models. The performance of the model was evaluated by the area under the subject working characteristic curve (AUC). The important predictive factors are determined by ranking the characteristic importance of the predictive factors. A total of 17 variables were screened; among the top 4 characteristics according to the weight coefficient in this model, urine WBC, urine occult blood, qualitative urinary protein, and microcyte percentage had high predictive value for kidney stones in patients. The accuracy of the kidney stone (KS-LSTM) learning model was 89.5%, and the AUC was 0.95. Compared with other models, it has better performance. The results show that the KS-LSTM model based on routine urine and blood tests can accurately identify the presence of kidney stones. And provide valuable assistance for clinicians to identify kidney stones in the early stage.

PMID:39402276 | DOI:10.1007/s00240-024-01644-6

NPJ Precis Oncol. 2024 Oct 14;8(1):232. doi: 10.1038/s41698-024-00690-y.

ABSTRACT

Early survival prediction is vital for the clinical management of cancer patients, as tumors can be better controlled with personalized treatment planning. Traditional survival prediction methods are based on radiomics feature engineering and/or clinical indicators (e.g., cancer staging). Recently, survival prediction models with advances in deep learning techniques have achieved state-of-the-art performance in end-to-end survival prediction by exploiting deep features derived from medical images. However, existing models are heavily reliant on the prognostic information within primary tumors and cannot effectively leverage out-of-tumor prognostic information characterizing local tumor metastasis and adjacent tissue invasion. Also, existing models are sub-optimal in leveraging multi-modality medical images as they rely on empirically designed fusion strategies to integrate multi-modality information, where the fusion strategies are pre-defined based on domain-specific human prior knowledge and inherently limited in adaptability. Here, we present an Adaptive Multi-modality Segmentation-to-Survival model (AdaMSS) for survival prediction from multi-modality medical images. The AdaMSS can self-adapt its fusion strategy based on training data and also can adapt its focus regions to capture the prognostic information outside the primary tumors. Extensive experiments with two large cancer datasets (1380 patients from nine medical centers) show that our AdaMSS surmounts the state-of-the-art survival prediction performance (C-index: 0.804 and 0.757), demonstrating the potential to facilitate personalized treatment planning.

PMID:39402129 | DOI:10.1038/s41698-024-00690-y

Sci Rep. 2024 Oct 14;14(1):23983. doi: 10.1038/s41598-024-74989-y.

ABSTRACT

Ensuring operational integrity in large-scale equipment hinges on effective fault prediction and health management. Prognostics and health management (PHM) face the challenge of accurately predicting remaining useful life (RUL) using multivariate sensor data. Traditional methods often require extensive prior knowledge for indicator construction and processing. Deep learning offers a promising alternative. This study presents a multi-channel multi-scale deep learning approach. Initially, an improved Savitzky‒Golay filter (ISG) addresses challenges posed by large and rapidly changing data volumes, enhancing data preprocessing. Subsequently, a framework integrates convolutional neural networks (CNNs) with long short-term memory (LSTM) to capture hierarchical signal information and make integrated predictions. The CNN extracts spatial features from multi-channel input data, while the LSTM captures temporal dependencies. By fusing outputs from both components, the framework enhances predictive accuracy and robustness for complex operational datasets. Experimental validation on the C-MAPSS dataset tests various fusion methods and CNN depths, determining parameters and evaluating filtering effectiveness. Comparative analyses show promising performance, particularly under dynamic conditions. While not optimal for predicting multiple fault types, it outperforms classical algorithms, especially in single fault type prediction tasks.

PMID:39402125 | DOI:10.1038/s41598-024-74989-y

Sci Rep. 2024 Oct 14;14(1):24064. doi: 10.1038/s41598-024-72427-7.

ABSTRACT

Malaria is a deadly disease caused by Plasmodium parasites. While potent drugs are available in the market for malaria treatment, over the years, Plasmodium parasites have successfully developed resistance against many, if not all, front-line drugs. This poses a serious threat to global malaria eradication efforts, and the continued discovery of new drugs is necessary to tackle this debilitating disease. With recent unprecedented progress in machine learning techniques, single-cell transcriptomic in Plasmodium offers a powerful tool for identifying crucial proteins as a drug target and subsequent computational prediction of potential drugs. In this study, We have implemented a mutual-information-based feature reduction algorithm with a classification algorithm to select important proteins from transcriptomic datasets (sexual and asexual stages) for Plasmodium falciparum and then constructed the protein-protein interaction (PPI) networks of the proteins. The analysis of this PPI network revealed key proteins vital for the survival of Plasmodium falciparum. Based on the function and identification of a few strong binding sites on a couple of these key proteins, we computationally predicted a set of potential drug molecules using a deep learning-based technique. Lead drug molecules that satisfy ADMET and drug-likeliness properties are finally reported out of the generated drugs. The study offers a general computational pipeline to identify crucial proteins using scRNA-seq data sets and further development of potential new drugs.

PMID:39402081 | DOI:10.1038/s41598-024-72427-7

Res Vet Sci. 2024 Oct 9;180:105434. doi: 10.1016/j.rvsc.2024.105434. Online ahead of print.

ABSTRACT

As a means of preventing mastitis, deep learning for classifying teat-end conditions in dairy cows has not yet been optimized. By using 1426 digital images of dairy cow udders, the extent of teat-end hyperkeratosis was assessed using a four-point scale. Several deep-learning networks based on the transfer learning approach have been used to evaluate the conditions of the teat ends displayed in the digital images. The images of the teat ends were partitioned into training (70 %) and validation datasets (15 %); afterwards, the network was evaluated based on the remaining test dataset (15 %). The results demonstrated that eight different ImageNet models consistently achieved high accuracy (80.3-86.6 %). The areas under the receiver operating characteristic curves for the normal, smooth, rough, and very rough classification scores in the test data set ranged from 0.825 to 0.999. Thus, improved accuracy in image-based classification of teat tissue conditions in dairy cattle using deep learning requires more training images. This method could help farmers reduce the risks of intramammary infections, decrease the use of antimicrobials, and better manage costs associated with mastitis detection and treatment.

PMID:39401476 | DOI:10.1016/j.rvsc.2024.105434

Med Phys. 2024 Oct 14. doi: 10.1002/mp.17459. Online ahead of print.

ABSTRACT

BACKGROUND: Single Photon Emission Computed Tomography (SPECT) sinogram restoration for low-dose imaging is a critical challenge in medical imaging. Existing methods often overlook the characteristics of the sinograms, necessitating innovative approaches.

PURPOSE: In this study, we introduce the Sinogram-characteristic-informed network (SCI-Net) to address the restoration of low-dose SPECT sinograms. Our aim is to build and train a model based on the characteristics of sinograms, including continuity, periodicity, multi-scale properties of lines in sinograms, and others, to enhance the model's understanding of the restoration process.

METHODS: SCI-Net incorporates several novel mechanisms tailored to exploit the inherent characteristics of sinograms. We implement a channel attention module with a decay mechanism to leverage continuity across adjacent sinograms, while a position attention module captures global correlations within individual sinograms. Additionally, we propose a multi-stage progressive integration mechanism to balance local detail and overall structure. Multiple regularization terms, customized to sinogram image characteristics, are embedded into the loss function for model training.

RESULTS: The experimental evaluations are divided into two parts: simulation data evaluation and clinical evaluation. The simulation data evaluation is conducted on a dataset comprising ten organ types, generated by the SIMIND Monte Carlo program from extended cardiac-torso (XCAT) anatomical phantoms. The dataset includes a total of SPECT sinograms with low-dose as input data and normal-dose as ground truth, consisting of 3881 sinograms in the training dataset and 849 sinograms in the testing set. When comparing the restoration of low-dose sinograms to normal-dose references, SCI-Net effectively improves performance. Specifically, the peak signal-to-noise ratio (PSNR) and the structural similarity (SSIM) on sinograms increase from 15.72 to 34.66 ( p < $\text{p}&lt; $ 0.001) and 0.6297 to 0.9834 ( p < $\text{p}&lt;$ 0.001), respectively, and on reconstructed images, reconstructed by maximum likelihood-expectation maximization (ML-EM), the PSNR and the SSIM improve from 21.95 to 33.14 ( p < $\text{p}&lt;$ 0.001) and 0.9084 to 0.9866 ( p < $\text{p}&lt;$ 0.001), respectively. We compared SCI-Net with existing methods, including baseline models, traditional reconstruction algorithms, end-to-end methods, sinogram restoration methods, and image post-processing methods. The experimental results and visual examples demonstrate that SCI-Net surpasses these existing methods in SPECT sinogram restoration. The clinical evaluation is conducted on clinical data of low-dose SPECT sinograms for spleen, thyroid, skull, and bone. These SPECT projection data are obtained from Discovery NM/CT670 scans. We compare the reconstructed images from the SCI-Net restored sinograms, the reconstructed images from the original low-dose sinograms, and the reconstructed images using the built-in algorithm of the Discovery NM/CT670. The results show that our method effectively reduces the coefficient of variation (COV) in the regions of interest (ROI) of the reconstructed images, thereby enhancing the quality of the reconstructed images through SPECT sinogram restoration.

CONCLUSIONS: Our proposed SCI-Net exhibits promising performance in the restoration of low-dose SPECT projection data. In the SCI-Net, we have implemented three mechanisms based on distinct forms, which are advantageous for the model to more effectively leverage the characteristics of sinograms and achieve commendable restoration outcomes.

PMID:39401269 | DOI:10.1002/mp.17459

J Appl Clin Med Phys. 2024 Oct 14:e14553. doi: 10.1002/acm2.14553. Online ahead of print.

ABSTRACT

PURPOSE: To explore the accuracy and feasibility of a proposed deep learning (DL) algorithm for clinical target volume (CTV) delineation in cervical cancer radiotherapy and evaluate whether it can perform well in external cervical cancer and endometrial cancer cases for generalization validation.

METHODS: A total of 332 patients were enrolled in this study. A state-of-the-art network called ResCANet, which added the cascade multi-scale convolution in the skip connections to eliminate semantic differences between different feature layers based on ResNet-UNet. The atrous spatial pyramid pooling in the deepest feature layer combined the semantic information of different receptive fields without losing information. A total of 236 cervical cancer cases were randomly grouped into 5-fold cross-training (n = 189) and validation (n = 47) cohorts. External validations were performed in a separate cohort of 54 cervical cancer and 42 endometrial cancer cases. The performances of the proposed network were evaluated by dice similarity coefficient (DSC), sensitivity (SEN), positive predictive value (PPV), 95% Hausdorff distance (95HD), and oncologist clinical score when comparing them with manual delineation in validation cohorts.

RESULTS: In internal validation cohorts, the mean DSC, SEN, PPV, 95HD for ResCANet achieved 74.8%, 81.5%, 73.5%, and 10.5 mm. In external independent validation cohorts, ResCANet achieved 73.4%, 72.9%, 75.3%, 12.5 mm for cervical cancer cases and 77.1%, 81.1%, 75.5%, 10.3 mm for endometrial cancer cases, respectively. The clinical assessment score showed that minor and no revisions (delineation time was shortened to within 30 min) accounted for about 85% of all cases in DL-aided automatic delineation.

CONCLUSIONS: We demonstrated the problem of model generalizability for DL-based automatic delineation. The proposed network can improve the performance of automatic delineation for cervical cancer and shorten manual delineation time at no expense to quality. The network showed excellent clinical viability, which can also be even generalized for endometrial cancer with excellent performance.

PMID:39401180 | DOI:10.1002/acm2.14553

Brief Bioinform. 2024 Sep 23;25(6):bbae504. doi: 10.1093/bib/bbae504.

ABSTRACT

Subcellular localization of messenger ribonucleic acid (mRNA) is a universal mechanism for precise and efficient control of the translation process. Although many computational methods have been constructed by researchers for predicting mRNA subcellular localization, very few of these computational methods have been designed to predict subcellular localization with multiple localization annotations, and their generalization performance could be improved. In this study, the prediction model MSlocPRED was constructed to identify multi-label mRNA subcellular localization. First, the preprocessed Dataset 1 and Dataset 2 are transformed into the form of images. The proposed MDNDO-SMDU resampling technique is then used to balance the number of samples in each category in the training dataset. Finally, deep transfer learning was used to construct the predictive model MSlocPRED to identify subcellular localization for 16 classes (Dataset 1) and 18 classes (Dataset 2). The results of comparative tests of different resampling techniques show that the resampling technique proposed in this study is more effective in preprocessing for subcellular localization. The prediction results of the datasets constructed by intercepting different NC end (Both the 5' and 3' untranslated regions that flank the protein-coding sequence and influence mRNA function without encoding proteins themselves.) lengths show that for Dataset 1 and Dataset 2, the prediction performance is best when the NC end is intercepted by 35 nucleotides, respectively. The results of both independent testing and five-fold cross-validation comparisons with established prediction tools show that MSlocPRED is significantly better than established tools for identifying multi-label mRNA subcellular localization. Additionally, to understand how the MSlocPRED model works during the prediction process, SHapley Additive exPlanations was used to explain it. The predictive model and associated datasets are available on the following github: https://github.com/ZBYnb1/MSlocPRED/tree/main.

PMID:39401145 | DOI:10.1093/bib/bbae504

IEEE J Biomed Health Inform. 2024 Oct 14;PP. doi: 10.1109/JBHI.2024.3480197. Online ahead of print.

ABSTRACT

Fetal multi-anatomical structure detection in Ultrasound (US) images can clearly present the relationship and influence between anatomical structures, providing more comprehensive information about fetal organ structures and assisting sonographers in making more accurate diagnoses, widely used in structure evaluation. Recently, deep learning methods have shown superior performance in detecting various anatomical structures in ultrasound images but still have the potential for performance improvement in categories where it is difficult to obtain samples, such as rare diseases. Few-shot learning has attracted a lot of attention in medical image analysis due to its ability to solve the problem of data scarcity. However, existing few-shot learning research in medical image analysis focuses on classification and segmentation, and the research on object detection has been neglected. In this paper, we propose a novel fetal anatomical structure fewshot detection method in ultrasound images, TKR-FSOD, which learns topological knowledge through a Topological Knowledge Reasoning Module to help the model reason about and detect anatomical structures. Furthermore, we propose a Discriminate Ability Enhanced Feature Learning Module that extracts abundant discriminative features to enhance the model's discriminative ability. Experimental results demonstrate that our method outperforms the state-of-the-art baseline methods, exceeding the second best method with a maximum margin of 4.8% on 5-shot of split 1 under 4CC. The code is publicly available at: https://github.com/lixi92/TKR-FSOD.

PMID:39401118 | DOI:10.1109/JBHI.2024.3480197

IEEE/ACM Trans Comput Biol Bioinform. 2024 Oct 14;PP. doi: 10.1109/TCBB.2024.3480088. Online ahead of print.

ABSTRACT

Currently, biomedical event extraction has received considerable attention in various fields, including natural language processing, bioinformatics, and computational biomedicine. This has led to the emergence of numerous machine learning and deep learning models that have been proposed and applied to tackle this complex task. While existing models typically adopt an extraction-based approach, which requires breaking down the extraction of biomedical events into multiple subtasks for sequential processing, making it prone to cascading errors. This paper presents a novel approach by constructing a biomedical event generation model based on the framework of the pre-trained language model T5. We employ a sequence-tosequence generation paradigm to obtain events, the model utilizes constrained decoding algorithm to guide sequence generation, and a curriculum learning algorithm for efficient model learning. To demonstrate the effectiveness of our model, we evaluate it on two public benchmark datasets, Genia 2011 and Genia 2013. Our model achieves superior performance, illustrating the effectiveness of generative modeling of biomedical events.

PMID:39401115 | DOI:10.1109/TCBB.2024.3480088

Small Methods. 2024 Oct 14:e2400672. doi: 10.1002/smtd.202400672. Online ahead of print.

ABSTRACT

Generative models, such as diffusion models, have made significant advancements in recent years, enabling the synthesis of high-quality realistic data across various domains. Here, the adaptation and training of a diffusion model on super-resolution microscopy images are explored. It is shown that the generated images resemble experimental images, and that the generation process does not exhibit a large degree of memorization from existing images in the training set. To demonstrate the usefulness of the generative model for data augmentation, the performance of a deep learning-based single-image super-resolution (SISR) method trained using generated high-resolution data is compared against training using experimental images alone, or images generated by mathematical modeling. Using a few experimental images, the reconstruction quality and the spatial resolution of the reconstructed images are improved, showcasing the potential of diffusion model image generation for overcoming the limitations accompanying the collection and annotation of microscopy images. Finally, the pipeline is made publicly available, runnable online, and user-friendly to enable researchers to generate their own synthetic microscopy data. This work demonstrates the potential contribution of generative diffusion models for microscopy tasks and paves the way for their future application in this field.

PMID:39400948 | DOI:10.1002/smtd.202400672