Future Med Chem. 2024 Aug 15:1-17. doi: 10.1080/17568919.2024.2385288. Online ahead of print.

ABSTRACT

Aim: Build a virtual screening model for ULK1 inhibitors based on artificial intelligence. Materials & methods: Build machine learning and deep learning classification models and combine molecular docking and biological evaluation to screen ULK1 inhibitors from 13 million compounds. And molecular dynamics was used to explore the binding mechanism of active compounds. Results & conclusion: Possibly due to less available training data, machine learning models significantly outperform deep learning models. Among them, the Naive Bayes model has the best performance. Through virtual screening, we obtained three inhibitors with IC50 of μM level and they all bind well to ULK1. This study provides an efficient virtual screening model and three promising compounds for the study of ULK1 inhibitors.

PMID:39145469 | DOI:10.1080/17568919.2024.2385288

J Clin Invest. 2024 Aug 15;134(16):e172885. doi: 10.1172/JCI172885.

ABSTRACT

Alcohol use disorder (AUD) is a prominent contributor to global morbidity and mortality. Its complex etiology involves genetics, epigenetics, and environmental factors. We review progress in understanding the genetics and epigenetics of AUD, summarizing the key findings. Advancements in technology over the decades have elevated research from early candidate gene studies to present-day genome-wide scans, unveiling numerous genetic and epigenetic risk factors for AUD. The latest GWAS on more than one million participants identified more than 100 genetic variants, and the largest epigenome-wide association studies (EWAS) in blood and brain samples have revealed tissue-specific epigenetic changes. Downstream analyses revealed enriched pathways, genetic correlations with other traits, transcriptome-wide association in brain tissues, and drug-gene interactions for AUD. We also discuss limitations and future directions, including increasing the power of GWAS and EWAS studies as well as expanding the diversity of populations included in these analyses. Larger samples, novel technologies, and analytic approaches are essential; these include whole-genome sequencing, multiomics, single-cell sequencing, spatial transcriptomics, deep-learning prediction of variant function, and integrated methods for disease risk prediction.

PMID:39145449 | DOI:10.1172/JCI172885

Stroke. 2024 Aug 15. doi: 10.1161/STROKEAHA.124.047449. Online ahead of print.

ABSTRACT

BACKGROUND: Cerebral small vessel disease is the most common pathology underlying vascular dementia. In small vessel disease, diffusion tensor imaging is more sensitive to white matter damage and better predicts dementia risk than conventional magnetic resonance imaging sequences, such as T1 and fluid attenuation inversion recovery, but diffusion tensor imaging takes longer to acquire and is not routinely available in clinical practice. As diffusion tensor imaging-derived scalar maps-fractional anisotropy (FA) and mean diffusivity (MD)-are frequently used in clinical settings, one solution is to synthesize FA/MD from T1 images.

METHODS: We developed a deep learning model to synthesize FA/MD from T1. The training data set consisted of 4998 participants with the highest white matter hyperintensity volumes in the UK Biobank. Four external validations data sets with small vessel disease were included: SCANS (St George's Cognition and Neuroimaging in Stroke; n=120), RUN DMC (Radboud University Nijmegen Diffusion Tensor and Magnetic Resonance Imaging Cohort; n=502), PRESERVE (Blood Pressure in Established Cerebral Small Vessel Disease; n=105), and NETWORKS (n=26), along with 1000 normal controls from the UK Biobank.

RESULTS: The synthetic maps resembled ground-truth maps (structural similarity index >0.89 for MD maps and >0.80 for FA maps across all external validation data sets except for SCANS). The prediction accuracy of dementia using whole-brain median MD from the synthetic maps is comparable to the ground truth (SCANS ground-truth c-index, 0.822 and synthetic, 0.821; RUN DMC ground truth, 0.816 and synthetic, 0.812) and better than white matter hyperintensity volume (SCANS, 0.534; RUN DMC, 0.710).

CONCLUSIONS: We have developed a fast and generalizable method to synthesize FA/MD maps from T1 to improve the prediction accuracy of dementia in small vessel disease when diffusion tensor imaging data have not been acquired.

PMID:39145386 | DOI:10.1161/STROKEAHA.124.047449

PeerJ Comput Sci. 2024 Jul 19;10:e2171. doi: 10.7717/peerj-cs.2171. eCollection 2024.

ABSTRACT

BACKGROUND: Cancer remains one of the leading causes of mortality globally, with conventional chemotherapy often resulting in severe side effects and limited effectiveness. Recent advancements in bioinformatics and machine learning, particularly deep learning, offer promising new avenues for cancer treatment through the prediction and identification of anticancer peptides.

OBJECTIVE: This study aimed to develop and evaluate a deep learning model utilizing a two-dimensional convolutional neural network (2D CNN) to enhance the prediction accuracy of anticancer peptides, addressing the complexities and limitations of current prediction methods.

METHODS: A diverse dataset of peptide sequences with annotated anticancer activity labels was compiled from various public databases and experimental studies. The sequences were preprocessed and encoded using one-hot encoding and additional physicochemical properties. The 2D CNN model was trained and optimized using this dataset, with performance evaluated through metrics such as accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC).

RESULTS: The proposed 2D CNN model achieved superior performance compared to existing methods, with an accuracy of 0.87, precision of 0.85, recall of 0.89, F1-score of 0.87, and an AUC-ROC value of 0.91. These results indicate the model's effectiveness in accurately predicting anticancer peptides and capturing intricate spatial patterns within peptide sequences.

CONCLUSION: The findings demonstrate the potential of deep learning, specifically 2D CNNs, in advancing the prediction of anticancer peptides. The proposed model significantly improves prediction accuracy, offering a valuable tool for identifying effective peptide candidates for cancer treatment.

FUTURE WORK: Further research should focus on expanding the dataset, exploring alternative deep learning architectures, and validating the model's predictions through experimental studies. Efforts should also aim at optimizing computational efficiency and translating these predictions into clinical applications.

PMID:39145253 | PMC:PMC11323142 | DOI:10.7717/peerj-cs.2171

PeerJ Comput Sci. 2024 Aug 7;10:e2222. doi: 10.7717/peerj-cs.2222. eCollection 2024.

ABSTRACT

Natural language processing (NLP) tasks can be addressed with several deep learning architectures, and many different approaches have proven to be efficient. This study aims to briefly summarize the use cases for NLP tasks along with the main architectures. This research presents transformer-based solutions for NLP tasks such as Bidirectional Encoder Representations from Transformers (BERT), and Generative Pre-Training (GPT) architectures. To achieve that, we conducted a step-by-step process in the review strategy: identify the recent studies that include Transformers, apply filters to extract the most consistent studies, identify and define inclusion and exclusion criteria, assess the strategy proposed in each study, and finally discuss the methods and architectures presented in the resulting articles. These steps facilitated the systematic summarization and comparative analysis of NLP applications based on Transformer architectures. The primary focus is the current state of the NLP domain, particularly regarding its applications, language models, and data set types. The results provide insights into the challenges encountered in this research domain.

PMID:39145251 | PMC:PMC11322986 | DOI:10.7717/peerj-cs.2222

PeerJ Comput Sci. 2024 Jul 17;10:e2189. doi: 10.7717/peerj-cs.2189. eCollection 2024.

ABSTRACT

Self-localization and pose registration are required for sound operation of next generation autonomous vehicles under uncertain environments. Thus, precise localization and mapping are crucial tasks in odometry, planning and other downstream processing. In order to reduce information loss in preprocessing, we propose leveraging LiDAR-based localization and mapping (LOAM) with point cloud-based deep learning instead of convolutional neural network (CNN) based methods that require cylindrical projection. The normal distribution transform (NDT) algorithm is then used to refine the former coarse pose estimation from the deep learning model. The results demonstrate that the proposed method is comparable in performance to recent benchmark studies. We also explore the possibility of using Product Quantization to improve NDT internal neighborhood searching by using high-level features as fingerprints.

PMID:39145248 | PMC:PMC11322985 | DOI:10.7717/peerj-cs.2189

PeerJ Comput Sci. 2024 Aug 7;10:e2193. doi: 10.7717/peerj-cs.2193. eCollection 2024.

ABSTRACT

The combination of memory forensics and deep learning for malware detection has achieved certain progress, but most existing methods convert process dump to images for classification, which is still based on process byte feature classification. After the malware is loaded into memory, the original byte features will change. Compared with byte features, function call features can represent the behaviors of malware more robustly. Therefore, this article proposes the ProcGCN model, a deep learning model based on DGCNN (Deep Graph Convolutional Neural Network), to detect malicious processes in memory images. First, the process dump is extracted from the whole system memory image; then, the Function Call Graph (FCG) of the process is extracted, and feature vectors for the function node in the FCG are generated based on the word bag model; finally, the FCG is input to the ProcGCN model for classification and detection. Using a public dataset for experiments, the ProcGCN model achieved an accuracy of 98.44% and an F1 score of 0.9828. It shows a better result than the existing deep learning methods based on static features, and its detection speed is faster, which demonstrates the effectiveness of the method based on function call features and graph representation learning in memory forensics.

PMID:39145247 | PMC:PMC11323106 | DOI:10.7717/peerj-cs.2193

PeerJ Comput Sci. 2024 Jul 10;10:e2115. doi: 10.7717/peerj-cs.2115. eCollection 2024.

ABSTRACT

In today's digital world, app stores have become an essential part of software distribution, providing customers with a wide range of applications and opportunities for software developers to showcase their work. This study elaborates on the importance of end-user feedback for software evolution. However, in the literature, more emphasis has been given to high-rating & popular software apps while ignoring comparatively low-rating apps. Therefore, the proposed approach focuses on end-user reviews collected from 64 low-rated apps representing 14 categories in the Amazon App Store. We critically analyze feedback from low-rating apps and developed a grounded theory to identify various concepts important for software evolution and improving its quality including user interface (UI) and user experience (UX), functionality and features, compatibility and device-specific, performance and stability, customer support and responsiveness and security and privacy issues. Then, using a grounded theory and content analysis approach, a novel research dataset is curated to evaluate the performance of baseline machine learning (ML), and state-of-the-art deep learning (DL) algorithms in automatically classifying end-user feedback into frequently occurring issues. Various natural language processing and feature engineering techniques are utilized for improving and optimizing the performance of ML and DL classifiers. Also, an experimental study comparing various ML and DL algorithms, including multinomial naive Bayes (MNB), logistic regression (LR), random forest (RF), multi-layer perception (MLP), k-nearest neighbors (KNN), AdaBoost, Voting, convolutional neural network (CNN), long short-term memory (LSTM), bidirectional long short term memory (BiLSTM), gated recurrent unit (GRU), bidirectional gated recurrent unit (BiGRU), and recurrent neural network (RNN) classifiers, achieved satisfactory results in classifying end-user feedback to commonly occurring issues. Whereas, MLP, RF, BiGRU, GRU, CNN, LSTM, and Classifiers achieved average accuracies of 94%, 94%, 92%, 91%, 90%, 89%, and 89%, respectively. We employed the SHAP approach to identify the critical features associated with each issue type to enhance the explainability of the classifiers. This research sheds light on areas needing improvement in low-rated apps and opens up new avenues for developers to improve software quality based on user feedback.

PMID:39145243 | PMC:PMC11323132 | DOI:10.7717/peerj-cs.2115

PeerJ Comput Sci. 2024 Jul 19;10:e2150. doi: 10.7717/peerj-cs.2150. eCollection 2024.

ABSTRACT

Virtual reality (VR) and immersive technology have emerged as powerful tools with numerous applications. VR technology creates a computer-generated simulation that immerses users in a virtual environment, providing a highly realistic and interactive experience. This technology finds applications in various fields, including gaming, healthcare, education, architecture, and training simulations. Understanding user immersion levels in VR is crucial and challenging for optimizing the design of VR applications. Immersion refers to the extent to which users feel absorbed and engrossed in the virtual environment. This research primarily aims to detect user immersion levels in VR using an efficient machine-learning model. We utilized a benchmark dataset based on user experiences in VR environments to conduct our experiments. Advanced deep and machine learning approaches are applied in comparison. We proposed a novel technique called Polynomial Random Forest (PRF) for feature generation mechanisms. The proposed PRF approach extracts polynomial and class prediction probability features to generate a new feature set. Extensive research experiments show that random forest outperformed state-of-the-art approaches, achieving a high immersion level detection rate of 98%, using the proposed PRF technique. We applied hyperparameter optimization and cross-validation approaches to validate the performance scores. Additionally, we utilized explainable artificial intelligence (XAI) to interpret the reasoning behind the decisions made by the proposed model for user immersion level detection in VR. Our research has the potential to revolutionize user immersion level detection in VR, enhancing the design process.

PMID:39145242 | PMC:PMC11323078 | DOI:10.7717/peerj-cs.2150

PeerJ Comput Sci. 2024 Jul 11;10:e2124. doi: 10.7717/peerj-cs.2124. eCollection 2024.

ABSTRACT

Pashtu is one of the most widely spoken languages in south-east Asia. Pashtu Numerics recognition poses challenges due to its cursive nature. Despite this, employing a machine learning-based optical character recognition (OCR) model can be an effective way to tackle this issue. The main aim of the study is to propose an optimized machine learning model which can efficiently identify Pashtu numerics from 0-9. The methodology includes data organizing into different directories each representing labels. After that, the data is preprocessed i.e., images are resized to 32 × 32 images, then they are normalized by dividing their pixel value by 255, and the data is reshaped for model input. The dataset was split in the ratio of 80:20. After this, optimized hyperparameters were selected for LSTM and CNN models with the help of trial-and-error technique. Models were evaluated by accuracy and loss graphs, classification report, and confusion matrix. The results indicate that the proposed LSTM model slightly outperforms the proposed CNN model with a macro-average of precision: 0.9877, recall: 0.9876, F1 score: 0.9876. Both models demonstrate remarkable performance in accurately recognizing Pashtu numerics, achieving an accuracy level of nearly 98%. Notably, the LSTM model exhibits a marginal advantage over the CNN model in this regard.

PMID:39145239 | PMC:PMC11323096 | DOI:10.7717/peerj-cs.2124

PeerJ Comput Sci. 2024 Aug 8;10:e2169. doi: 10.7717/peerj-cs.2169. eCollection 2024.

ABSTRACT

The Boolean satisfiability (SAT) problem exhibits different structural features in various domains. Neural network models can be used as more generalized algorithms that can be learned to solve specific problems based on different domain data than traditional rule-based approaches. How to accurately identify these structural features is crucial for neural networks to solve the SAT problem. Currently, learning-based SAT solvers, whether they are end-to-end models or enhancements to traditional heuristic algorithms, have achieved significant progress. In this article, we propose TG-SAT, an end-to-end framework based on Transformer and gated recurrent neural network (GRU) for predicting the satisfiability of SAT problems. TG-SAT can learn the structural features of SAT problems in a weakly supervised environment. To capture the structural information of the SAT problem, we encodes a SAT problem as an undirected graph and integrates GRU into the Transformer structure to update the node embeddings. By computing cross-attention scores between literals and clauses, a weighted representation of nodes is obtained. The model is eventually trained as a classifier to predict the satisfiability of the SAT problem. Experimental results demonstrate that TG-SAT achieves a 2%-5% improvement in accuracy on random 3-SAT problems compared to NeuroSAT. It also outperforms in SR(N), especially in handling more complex SAT problems, where our model achieves higher prediction accuracy.

PMID:39145235 | PMC:PMC11323027 | DOI:10.7717/peerj-cs.2169

PeerJ Comput Sci. 2024 Jul 23;10:e2009. doi: 10.7717/peerj-cs.2009. eCollection 2024.

ABSTRACT

With the popularity of Internet applications, a large amount of Internet behavior log data is generated. Abnormal behaviors of corporate employees may lead to internet security issues and data leakage incidents. To ensure the safety of information systems, it is important to research on anomaly prediction of Internet behaviors. Due to the high cost of labeling big data manually, an unsupervised generative model-Anomaly Prediction of Internet behavior based on Generative Adversarial Networks (APIBGAN), which works only with a small amount of labeled data, is proposed to predict anomalies of Internet behaviors. After the input Internet behavior data is preprocessed by the proposed method, the data-generating generative adversarial network (DGGAN) in APIBGAN learns the distribution of real Internet behavior data by leveraging neural networks' powerful feature extraction from the data to generate Internet behavior data with random noise. The APIBGAN utilizes these labeled generated data as a benchmark to complete the distance-based anomaly prediction. Three categories of Internet behavior sampling data from corporate employees are employed to train APIBGAN: (1) Online behavior data of an individual in a department. (2) Online behavior data of multiple employees in the same department. (3) Online behavior data of multiple employees in different departments. The prediction scores of the three categories of Internet behavior data are 87.23%, 85.13%, and 83.47%, respectively, and are above the highest score of 81.35% which is obtained by the comparison method based on Isolation Forests in the CCF Big Data & Computing Intelligence Contest (CCF-BDCI). The experimental results validate that APIBGAN predicts the outlier of Internet behaviors effectively through the GAN, which is composed of a simple three-layer fully connected neural networks (FNNs). We can use APIBGAN not only for anomaly prediction of Internet behaviors but also for anomaly prediction in many other applications, which have big data infeasible to label manually. Above all, APIBGAN has broad application prospects for anomaly prediction, and our work also provides valuable input for anomaly prediction-based GAN.

PMID:39145230 | PMC:PMC11323085 | DOI:10.7717/peerj-cs.2009

J Cardiovasc Magn Reson. 2024 Aug 12:101082. doi: 10.1016/j.jocmr.2024.101082. Online ahead of print.

ABSTRACT

BACKGROUND: Fully automatic analysis of myocardial perfusion MRI datasets enables rapid and objective reporting of stress/rest studies in patients with suspected ischemic heart disease. Developing deep learning techniques that can analyze multi-center datasets despite limited training data and variations in software (pulse sequence) and hardware (scanner vendor) is an ongoing challenge.

METHODS: Datasets from 3 medical centers acquired at 3T (n = 150 subjects; 21,150 first-pass images) were included: an internal dataset (inD; n = 95) and two external datasets (exDs; n = 55) used for evaluating the robustness of the trained deep neural network (DNN) models against differences in pulse sequence (exD-1) and scanner vendor (exD-2). A subset of inD (n = 85) was used for training/validation of a pool of DNNs for segmentation, all using the same spatiotemporal U-Net architecture and hyperparameters but with different parameter initializations. We employed a space-time sliding-patch analysis approach that automatically yields a pixel-wise "uncertainty map" as a byproduct of the segmentation process. In our approach, dubbed Data Adaptive Uncertainty-Guided Space-time (DAUGS) analysis, a given test case is segmented by all members of the DNN pool and the resulting uncertainty maps are leveraged to automatically select the "best" one among the pool of solutions. For comparison, we also trained a DNN using the established approach with the same settings (hyperparameters, data augmentation, etc.).

RESULTS: The proposed DAUGS analysis approach performed similarly to the established approach on the internal dataset (Dice score for the testing subset of inD: 0.896 ± 0.050 vs. 0.890 ± 0.049; p = n.s.) whereas it significantly outperformed on the external datasets (Dice for exD-1: 0.885 ± 0.040 vs. 0.849 ± 0.065, p < 0.005; Dice for exD-2: 0.811 ± 0.070 vs. 0.728 ± 0.149, p < 0.005). Moreover, the number of image series with "failed" segmentation (defined as having myocardial contours that include bloodpool or are noncontiguous in ≥1 segment) was significantly lower for the proposed vs. the established approach (4.3% vs. 17.1%, p < 0.0005).

CONCLUSIONS: The proposed DAUGS analysis approach has the potential to improve the robustness of deep learning methods for segmentation of multi-center stress perfusion datasets with variations in the choice of pulse sequence, site location or scanner vendor.

PMID:39142567 | DOI:10.1016/j.jocmr.2024.101082

Biol Psychiatry Cogn Neurosci Neuroimaging. 2024 Aug 12:S2451-9022(24)00237-4. doi: 10.1016/j.bpsc.2024.07.027. Online ahead of print.

ABSTRACT

BACKGROUND: Psychiatric disorders are traditionally classified within diagnostic categories, but this approach has limitations. Research Domain Criteria (RDoC) constitute a research classification system for psychiatric disorders based on dimensions within domains that cut across these psychiatric diagnoses. The overall aim of RDoC is to better understand mental illness in terms of dysfunction in fundamental neurobiological and behavioral systems, leading to better diagnosis, prevention and treatment.

METHODS: A unique electroencephalographic (EEG) feature, referred to as spindling excessive beta (SEB), has been studied in relation to impulse control and sleep, as part of the arousal/regulatory systems RDoC domain. Here, we study EEG frontal beta activity as a potential transdiagnostic biomarker capable of diagnosing and predicting impulse control and sleep problems.

RESULTS: We show in the first dataset (n=3279) that the probability of having SEB, classified by a deep learning algorithm, is associated with poor sleep maintenance and low daytime impulse control. Furthermore, in two additional, independent datasets (iSPOT-A, n=336; iSPOT-D, n=1008), we revealed that conventional frontocentral beta power and/or SEB probability, referred to as Brainmarker-III, is associated with a diagnosis of attention deficit hyperactivity disorder (ADHD), with remission to methylphenidate in children with ADHD in a sex-specific manner, and with remission to antidepressant medication in adults with a major depressive disorder in a drug-specific manner.

CONCLUSION: Our results demonstrate the value of the RDoC approach in psychiatry research for the discovery of biomarkers with diagnostic and treatment prediction capacities.

PMID:39142534 | DOI:10.1016/j.bpsc.2024.07.027

Phys Med Biol. 2024 Aug 14. doi: 10.1088/1361-6560/ad6f6a. Online ahead of print.

ABSTRACT

Objective&#xD;Respiratory motion, cardiac motion, and inherently low signal-to-noise ratio (SNR) are major limitations of in vivo cardiac diffusion tensor imaging (DTI). We propose a novel enhancement method that uses unsupervised learning based invertible wavelet scattering (IWS) to improve the quality of in vivo cardiac DTI.&#xD;&#xD;Approach&#xD;Our method starts by extracting nearly transformation-invariant features from multiple cardiac diffusion-weighted (DW) image acquisitions using multi-scale wavelet scattering (WS). The relationship between the WS coefficients and DW images is learned through a multiscale encoder and a decoder network. Using the trained encoder, the deep features of WS coefficients of multiple DW image acquisitions are further extracted and then fused using an average rule. Finally, using the fused WS features and trained decoder, the enhanced DW images are derived.&#xD;&#xD;Main Results&#xD;We evaluated the performance of the proposed method by comparing it with several methods on three in vivo cardiac DTI datasets in terms of SNR, contrast to noise ratio (CNR), fractional anisotropy (FA), mean diffusivity (MD), and helix angle (HA). Compared to the best comparison method, SNR/CNR of diastolic, gastric peristalsis influenced, and end systolic DW images were improved by 1%/16%, 5%/6%, and 56%/30%, respectively. The approach also yielded consistent FA and MD values and more coherent helical fiber structures than the comparison methods used in this work.&#xD;&#xD;Significance&#xD;The ablation results verify that using the transformation-invariant and noise-robust wavelet scattering features enables effective exploration of useful information from limited data. This provides a potential means to alleviate the dependence of the fusion results on the number of repeated acquisitions, which is beneficial for dealing with noise and residual motion issues simultaneously, thereby improving the quality of in vivo cardiac DTI.

PMID:39142339 | DOI:10.1088/1361-6560/ad6f6a

Biomed Phys Eng Express. 2024 Aug 14. doi: 10.1088/2057-1976/ad6f17. Online ahead of print.

ABSTRACT

Neuromyelitis optica spectrum disorder (NMOSD), also known as Devic disease, is an autoimmune central nervous system disorder in humans that commonly causes inflammatory demyelination in the optic nerves and spinal cord. Inflammation in the optic nerves is termed optic neuritis (ON). ON is a common clinical presentation; however, it is not necessarily present in all NMOSD patients. ON in NMOSD can be relapsing and result in severe vision loss. To the best of our knowledge, no study utilises deep learning to classify ON changes on MRI among patients with NMOSD. Therefore, this study aims to deploy eight state-of-the-art CNN models (Inception-v3, Inception-ResNet-v2, ResNet-101, Xception, ShuffleNet, DenseNet-201, MobileNet-v2, and EfficientNet-B0) with transfer learning to classify NMOSD patients with and without chronic ON using optic nerve magnetic resonance imaging. This study also investigated the effects of data augmentation before and after dataset splitting on cropped and whole images. Both quantitative and qualitative assessments (with Grad-Cam) were used to evaluate the performances of the CNN models. The Inception-v3 was identified as the best CNN model for classifying ON among NMOSD patients, with accuracy of 99.5%, sensitivity of 98.9%, specificity of 93.0%, precision of 100%, NPV of 99.0%, and F1-score of 99.4%. This study also demonstrated that the application of augmentation after dataset splitting could avoid information leaking into the testing datasets, hence producing more realistic and reliable results.

PMID:39142299 | DOI:10.1088/2057-1976/ad6f17

Biomed Phys Eng Express. 2024 Aug 14. doi: 10.1088/2057-1976/ad6f12. Online ahead of print.

ABSTRACT

With the advancement of computer-aided diagnosis, the automatic segmentation of COVID-19 infection areas holds great promise for assisting in the timely diagnosis and recovery of patients in clinical practice. Currently, methods relying on U-Net face challenges in effectively utilizing fine-grained semantic information from input images and bridging the semantic gap between the encoder and decoder. To address these issues, we propose an FMD-UNet dual-decoder U-Net network for COVID-19 infection segmentation, which integrates a Fine-grained Feature Squeezing (FGFS) decoder and a Multi-scale Dilated Semantic Aggregation (MDSA) decoder. The FGFS decoder produces fine feature maps through the compression of fine-grained features and a weighted attention mechanism, guiding the model to capture detailed semantic information. The MDSA decoder consists of three hierarchical MDSA modules designed for different stages of input information. These modules progressively fuse different scales of dilated convolutions to process the shallow and deep semantic information from the encoder, and use the extracted feature information to bridge the semantic gaps at various stages, this design captures extensive contextual information while decoding and predicting segmentation, thereby suppressing the increase in model parameters. To better validate the robustness and generalizability of the FMD-UNet, we conducted comprehensive performance evaluations and ablation experiments on three public datasets, and achieved leading Dice Similarity Coefficient (DSC) scores of 84.76, 78.56 and 61.99% in COVID-19 infection segmentation, respectively. Compared to previous methods, the FMD-UNet has fewer parameters and shorter inference time, which also demonstrates its competitiveness.

PMID:39142295 | DOI:10.1088/2057-1976/ad6f12

Comput Biol Med. 2024 Aug 13;180:108962. doi: 10.1016/j.compbiomed.2024.108962. Online ahead of print.

ABSTRACT

Today, doctors rely heavily on medical imaging to identify abnormalities. Proper classification of these abnormalities enables them to take informed actions, leading to early diagnosis and treatment. This paper introduces the "EfficientKNN" model, a novel hybrid deep learning approach that combines the advanced feature extraction capabilities of EfficientNetB3 with the simplicity and effectiveness of the k-Nearest Neighbors (k-NN) algorithm. Initially, EfficientNetB3, pre-trained on ImageNet, is repurposed to serve as a feature extractor. Subsequently, a GlobalAveragePooling2D layer is applied, followed by an optional Principal Component Analysis (PCA) to reduce dimensionality while preserving critical information. PCA is used selectively when deemed necessary. The extracted features are then classified using an optimized k-NN algorithm, fine-tuned through meticulous cross-validation.Our model underwent rigorous training using a curated dataset containing benign, malignant, and normal medical images. Data augmentation techniques, including rotations, shifts, flips, and zooms, were employed to help the model generalize and efficiently handle new, unseen data. To enhance the model's ability to identify the important features necessary for accurate predictions, the dataset was refined using segmentation and overlay techniques. The training utilized an ensemble of optimization algorithms-SGD, Adam, and RMSprop-with hyperparameters set at a learning rate of 0.00045, a batch size of 32, and up to 120 epochs, facilitated by early stopping to prevent overfitting.The results demonstrate that the EfficientKNN model outperforms traditional models such as VGG16, AlexNet, and VGG19 in terms of accuracy, precision, and F1-score. Additionally, the model showed better results compared to EfficientNetB3 alone. Achieving a 100 % accuracy rate on multiple tests, the EfficientKNN model has significant potential for real-world diagnostic applications. This study highlights the model's scalability, efficient use of cloud storage, and real-time prediction capabilities, all while minimizing computational demands.By integrating the strengths of EfficientNetB3's deep learning architecture with the interpretability of k-NN, EfficientKNN presents a significant advancement in medical image classification, promising improved diagnostic accuracy and clinical applicability.

PMID:39142222 | DOI:10.1016/j.compbiomed.2024.108962

Eur J Radiol. 2024 Aug 4;179:111663. doi: 10.1016/j.ejrad.2024.111663. Online ahead of print.

ABSTRACT

PURPOSE: To evaluate the impact of deep learning-based reconstruction (DLRecon) on bone assessment in zero echo-time (ZTE) MRI of the knee at 1.5 Tesla.

METHODS: This retrospective study included 48 consecutive exams of 46 patients (23 females) who underwent clinically indicated knee MRI at 1.5 Tesla. Standard imaging protocol comprised a sagittal prescribed, isotropic ZTE sequence. ZTE image reconstruction was performed with a standard-of-care (non-DL) and prototype DLRecon method. Exams were divided into subsets with and without osseous pathology based on the radiology report. Using a 4-point scale, two blinded readers qualitatively graded features of bone depiction including artifacts and conspicuity of pathology including diagnostic certainty in the respective subsets. Quantitatively, one reader measured signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of bone. Comparative analyses were conducted to assess the differences between the reconstruction methods. In addition, interreader agreement was calculated for the qualitative gradings.

RESULTS: DLRecon significantly improved gradings for bone depiction relative to non-DL reconstruction (all, p < 0.05), while there was no significant difference with regards to artifacts (both, median score of 0; p = 0.058). In the subset with pathologies, conspicuity of pathology and diagnostic confidence were also scored significantly higher in DLRecon compared to non-DL (median 3 vs 2; p ≤ 0.03). Interreader agreement ranged from moderate to almost-perfect (κ = 0.54-0.88). Quantitatively, DLRecon demonstrated significantly enhanced CNR and SNR of bone compared to non-DL (p < 0.001).

CONCLUSION: ZTE MRI with DLRecon improved bone depiction in the knee, compared to non-DL. Additionally, DLRecon increased conspicuity of osseous findings together with diagnostic certainty.

PMID:39142010 | DOI:10.1016/j.ejrad.2024.111663

J Transl Med. 2024 Aug 14;22(1):768. doi: 10.1186/s12967-024-05449-4.

ABSTRACT

BACKGROUND: Postoperative liver metastasis significantly impacts the prognosis of pancreatic neuroendocrine tumor (panNET) patients after R0 resection. Combining computational pathology and deep learning radiomics can enhance the detection of postoperative liver metastasis in panNET patients.

METHODS: Clinical data, pathology slides, and radiographic images were collected from 163 panNET patients post-R0 resection at Fudan University Shanghai Cancer Center (FUSCC) and FUSCC Pathology Consultation Center. Digital image analysis and deep learning identified liver metastasis-related features in Ki67-stained whole slide images (WSIs) and enhanced CT scans to create a nomogram. The model's performance was validated in both internal and external test cohorts.

RESULTS: Multivariate logistic regression identified nerve infiltration as an independent risk factor for liver metastasis (p < 0.05). The Pathomics score, which was based on a hotspot and the heterogeneous distribution of Ki67 staining, showed improved predictive accuracy for liver metastasis (AUC = 0.799). The deep learning-radiomics (DLR) score achieved an AUC of 0.875. The integrated nomogram, which combines clinical, pathological, and imaging features, demonstrated outstanding performance, with an AUC of 0.985 in the training cohort and 0.961 in the validation cohort. High-risk group had a median recurrence-free survival of 28.5 months compared to 34.7 months for the low-risk group, showing significant correlation with prognosis (p < 0.05).

CONCLUSION: A new predictive model that integrates computational pathologic scores and deep learning-radiomics can better predict postoperative liver metastasis in panNET patients, aiding clinicians in developing personalized treatments.

PMID:39143624 | DOI:10.1186/s12967-024-05449-4