JMIR Bioinform Biotechnol. 2024 Jul 24;5:e56538. doi: 10.2196/56538.

ABSTRACT

BACKGROUND: Carcinoma of unknown primary (CUP) is a subset of metastatic cancers in which the primary tissue source of the cancer cells remains unidentified. CUP is the eighth most common malignancy worldwide, accounting for up to 5% of all malignancies. Representing an exceptionally aggressive metastatic cancer, the median survival is approximately 3 to 6 months. The tissue in which cancer arises plays a key role in our understanding of sensitivities to various forms of cell death. Thus, the lack of knowledge on the tissue of origin (TOO) makes it difficult to devise tailored and effective treatments for patients with CUP. Developing quick and clinically implementable methods to identify the TOO of the primary site is crucial in treating patients with CUP. Noncoding RNAs may hold potential for origin identification and provide a robust route to clinical implementation due to their resistance against chemical degradation.

OBJECTIVE: This study aims to investigate the potential of microRNAs, a subset of noncoding RNAs, as highly accurate biomarkers for detecting the TOO through data-driven, machine learning approaches for metastatic cancers.

METHODS: We used microRNA expression data from The Cancer Genome Atlas data set and assessed various machine learning approaches, from simple classifiers to deep learning approaches. As a test of our classifiers, we evaluated the accuracy on a separate set of 194 primary tumor samples from the Sequence Read Archive. We used permutation feature importance to determine the potential microRNA biomarkers and assessed them with principal component analysis and t-distributed stochastic neighbor embedding visualizations.

RESULTS: Our results show that it is possible to design robust classifiers to detect the TOO for metastatic samples on The Cancer Genome Atlas data set, with an accuracy of up to 97% (351/362), which may be used in situations of CUP. Our findings show that deep learning techniques enhance prediction accuracy. We progressed from an initial accuracy prediction of 62.5% (226/362) with decision trees to 93.2% (337/362) with logistic regression, finally achieving 97% (351/362) accuracy using deep learning on metastatic samples. On the Sequence Read Archive validation set, a lower accuracy of 41.2% (77/188) was achieved by the decision tree, while deep learning achieved a higher accuracy of 80.4% (151/188). Notably, our feature importance analysis showed the top 3 most important features for predicting TOO to be microRNA-10b, microRNA-205, and microRNA-196b, which aligns with previous work.

CONCLUSIONS: Our findings highlight the potential of using machine learning techniques to devise accurate tests for detecting TOO for CUP. Since microRNAs are carried throughout the body via extracellular vesicles secreted from cells, they may serve as key biomarkers for liquid biopsy due to their presence in blood plasma. Our work serves as a foundation toward developing blood-based cancer detection tests based on the presence of microRNA.

PMID:39046787 | DOI:10.2196/56538

Syst Biol. 2024 Jul 24:syae042. doi: 10.1093/sysbio/syae042. Online ahead of print.

ABSTRACT

Reconstructing the tree of life and understanding the relationships of taxa are core questions in evolutionary and systematic biology. The main advances in this field in the last decades were derived from molecular phylogenetics; however, for most species, molecular data are not available. Here, we explore the applicability of two deep learning methods - supervised classification approaches and unsupervised similarity learning - to infer organism relationships from specimen images. As a basis, we assembled an image dataset covering 4144 bivalve species belonging to 74 families across all orders and subclasses of the extant Bivalvia, with molecular phylogenetic data being available for all families and a complete taxonomic hierarchy for all species. The suitability of this dataset for deep learning experiments was evidenced by an ablation study resulting in almost 80% accuracy for identifications on the species level. Three sets of experiments were performed using our dataset. First, we included taxonomic hierarchy and genetic distances in a supervised learning approach to obtain predictions on several taxonomic levels simultaneously. Here, we stimulated the model to consider features shared between closely related taxa to be more critical for their classification than features shared with distantly related taxa, imprinting phylogenetic and taxonomic affinities into the architecture and training procedure. Second, we used transfer learning and similarity learning approaches for zero-shot experiments to identify the higher-level taxonomic affinities of test species that the models had not been trained on. The models assigned the unknown species to their respective genera with approximately 48% and 67% accuracy. Lastly, we used unsupervised similarity learning to infer the relatedness of the images without prior knowledge of their taxonomic or phylogenetic affinities. The results clearly showed similarities between visual appearance and genetic relationships at the higher taxonomic levels. The correlation was 0.6 for the most species-rich subclass (Imparidentia), ranging from 0.5 to 0.7 for the orders with the most images. Overall, the correlation between visual similarity and genetic distances at the family level was 0.78. However, fine-grained reconstructions based on these observed correlations, such as sister-taxa relationships, require further work. Overall, our results broaden the applicability of automated taxon identification systems and provide a new avenue for estimating phylogenetic relationships from specimen images.

PMID:39046773 | DOI:10.1093/sysbio/syae042

Invest Ophthalmol Vis Sci. 2024 Jul 1;65(8):42. doi: 10.1167/iovs.65.8.42.

ABSTRACT

PURPOSE: AI algorithms have shown impressive performance in segmenting geographic atrophy (GA) from fundus autofluorescence (FAF) images. However, selection of artificial intelligence (AI) architecture is an important variable in model development. Here, we explore 12 distinct AI architecture combinations to determine the most effective approach for GA segmentation.

METHODS: We investigated various AI architectures, each with distinct combinations of encoders and decoders. The architectures included three decoders-FPN (Feature Pyramid Network), UNet, and PSPNet (Pyramid Scene Parsing Network)-and serve as the foundation framework for segmentation task. Encoders including EfficientNet, ResNet (Residual Networks), VGG (Visual Geometry Group) and Mix Vision Transformer (mViT) have a role in extracting optimum latent features for accurate GA segmentation. Performance was measured through comparison of GA areas between human and AI predictions and Dice Coefficient (DC).

RESULTS: The training dataset included 601 FAF images from AREDS2 study and validation included 156 FAF images from the GlaxoSmithKline study. The mean absolute difference between grader measured and AI predicted areas ranged from -0.08 (95% CI = -1.35, 1.19) to 0.73 mm2 (95% CI = -5.75,4.29) and DC between 0.884-0.993. The best-performing models were UNet and FPN frameworks with mViT, and the least-performing models were PSPNet framework.

CONCLUSIONS: The choice of AI architecture impacts GA segmentation performance. Vision transformers with FPN and UNet architectures demonstrate stronger suitability for this task compared to Convolutional Neural Network- and PSPNet-based models. Selecting an AI architecture must be tailored to the specific goals of the project, and developers should consider which architecture is ideal for their project.

PMID:39046755 | DOI:10.1167/iovs.65.8.42

Sleep Breath. 2024 Jul 24. doi: 10.1007/s11325-024-03112-2. Online ahead of print.

ABSTRACT

PURPOSE: The classification of sleep stages based on Electroencephalogram (EEG) changes has significant implications for evaluating sleep quality and sleep status. Most polysomnography (PSG) systems have a limited number of channels and do not achieve optimal classification performance due to a paucity of raw data. To leverage the data characteristics and enhance the classification accuracy, we propose and evaluate a novel dual-link deep neural network model, 'DoubleLinkSleepCLNet'.

METHODS: The DoubleLinkSleepCLNet model performs feature extraction and efficient classification on both the raw EEG and the EEG processed with the Hilbert transform. It leverages the frequency domain and time domain feature modules, resulting in superior performance compared to other models.

RESULTS: The DoubleLinkSleepCLNet model, using the 2 Raw/2 Hilbert data modes, achieved the highest classification performance with an accuracy of 88.47%. The average accuracy of the EEG was improved by approximately 4.08% after the application of the Hilbert transform. Additionally, Convolutional Neural Network (CNN) demonstrated superior performance in processing phase information, whereas Long Short-Term Memory (LSTM) excelled in handling time series data.

CONCLUSION: The application of the Hilbert transform to EEG data, followed by processing it with a convolutional neural network, enhances the accuracy of the model. These findings introduce novel concepts for accelerating sleep stage prediction research, suggesting potential applications of these methods to other EEG analyses.

PMID:39046659 | DOI:10.1007/s11325-024-03112-2

Environ Sci Pollut Res Int. 2024 Jul 24. doi: 10.1007/s11356-024-34404-5. Online ahead of print.

ABSTRACT

Hydrological simulation in karstic areas is a hard task due to the intrinsic intricacy of these environments and the common lack of data related to their geometry. Hydrological dynamics of karstic sites in Mediterranean semiarid regions are difficult to be modelled mathematically owing to the existence of short wet episodes and long dry periods. In this paper, the suitability of an open-source SWAT method was checked to estimate the comportment of a karstic catchment in a Mediterranean semiarid domain (southeast of Spain), which wet and dry periods were evaluated using box-whisker plots and self-developed wavelet test. A novel expression of the Nash-Sutcliffe index for arid areas (ANSE) was considered through the calibration and validation of SWAT. Both steps were completed with 20- and 10-year discharge records of stream (1996-2015 to calibrate the model as this period depicts minimum gaps and 1985-1995 to validate it). Further, SWAT assessments were made with records of groundwater discharge and relating SWAT outputs with the SIMPA method, the Spain's national hydrological tool. These methods, along with recurrent neural network algorithms, were utilised to examine current and predicted water resources available to supply urban demands considering also groundwater abstractions from aquifers and the related exploitation index. According to the results, SWAT achieved a "very good" statistical performance (with ANSE of 0.96 and 0.78 in calibration and validation). Spatial distributions of the main hydrological processes, as surface runoff, evapotranspiration and aquifer recharge, were studied with SWAT and SIMPA obtaining similar results over the period with registers (1980-2016). During this period, the decreasing trend of rainfalls, characterised by short wet periods and long dry periods, has generated a progressive reduction of groundwater recharge. According to algorithms prediction (until 2050), this declining trend will continue reducing groundwater available to meet urban demands and increasing the exploitation index of aquifers. These results offer valuable information to authorities for assessing water accessibility and to provide water demands in karstic areas.

PMID:39046638 | DOI:10.1007/s11356-024-34404-5

Endocrine. 2024 Jul 24. doi: 10.1007/s12020-024-03906-0. Online ahead of print.

ABSTRACT

PURPOSE: Thyroid eye disease (TED) is the most common orbital disease in adults. Ocular motility restriction is the primary complaint of patients, while its evaluation is quite difficult. The present study aimed to introduce an artificial intelligence (AI) model based on orbital computed tomography (CT) images for ocular motility score.

METHODS: A total of 410 sets of CT images and clinical data were obtained from the hospital. To build a triple classification predictive model for ocular motility score, multiple deep learning models were employed to extract features of images and clinical data. Subgroup analyses based on pertinent clinical features were performed to test the efficacy of models.

RESULTS: The ResNet-34 network outperformed Alex-Net and VGG16-Net in prediction of ocular motility score, with the optimal accuracy (ACC) of 0.907, 0.870, and 0.890, respectively. Subgroup analyses indicated no significant difference in ACC between active or inactive phase, functional visual field diplopia or peripheral visual field diplopia (p > 0.05). However, in the gender subgroup, the prediction model performed more accurately in female patients than males (p = 0.02).

CONCLUSION: In conclusion, the AI model based on CT images and clinical data successfully realized automatic scoring of ocular motility in TED patients. This approach potentially enhanced the efficiency and accuracy of ocular motility evaluation, thus facilitating clinical application.

PMID:39046593 | DOI:10.1007/s12020-024-03906-0

Eur Radiol Exp. 2024 Jul 24;8(1):84. doi: 10.1186/s41747-024-00486-6.

ABSTRACT

BACKGROUND: Computed tomography (CT) reconstruction algorithms can improve image quality, especially deep learning reconstruction (DLR). We compared DLR, iterative reconstruction (IR), and filtered back projection (FBP) for lesion detection in neck CT.

METHODS: Nine patient-mimicking neck phantoms were examined with a 320-slice scanner at six doses: 0.5, 1, 1.6, 2.1, 3.1, and 5.2 mGy. Each of eight phantoms contained one circular lesion (diameter 1 cm; contrast -30 HU to the background) in the parapharyngeal space; one phantom had no lesions. Reconstruction was made using FBP, IR, and DLR. Thirteen readers were tasked with identifying and localizing lesions in 32 images with a lesion and 20 without lesions for each dose and reconstruction algorithm. Receiver operating characteristic (ROC) and localization ROC (LROC) analysis were performed.

RESULTS: DLR improved lesion detection with ROC area under the curve (AUC) 0.724 ± 0.023 (mean ± standard error of the mean) using DLR versus 0.696 ± 0.021 using IR (p = 0.037) and 0.671 ± 0.023 using FBP (p < 0.001). Likewise, DLR improved lesion localization, with LROC AUC 0.407 ± 0.039 versus 0.338 ± 0.041 using IR (p = 0.002) and 0.313 ± 0.044 using FBP (p < 0.001). Dose reduction to 0.5 mGy compromised lesion detection in FBP-reconstructed images compared to doses ≥ 2.1 mGy (p ≤ 0.024), while no effect was observed with DLR or IR (p ≥ 0.058).

CONCLUSION: DLR improved the detectability of lesions in neck CT imaging. Dose reduction to 0.5 mGy maintained lesion detectability when denoising reconstruction was used.

RELEVANCE STATEMENT: Deep learning enhances lesion detection in neck CT imaging compared to iterative reconstruction and filtered back projection, offering improved diagnostic performance and potential for x-ray dose reduction.

KEY POINTS: Low-contrast lesion detectability was assessed in anatomically realistic neck CT phantoms. Deep learning reconstruction (DLR) outperformed filtered back projection and iterative reconstruction. Dose has little impact on lesion detectability against anatomical background structures.

PMID:39046565 | DOI:10.1186/s41747-024-00486-6

Pediatr Radiol. 2024 Jul 24. doi: 10.1007/s00247-024-05999-1. Online ahead of print.

ABSTRACT

BACKGROUND: Artificial intelligence has been increasingly used in medical imaging and has demonstrated expert level performance in image classification tasks.

OBJECTIVE: To develop a fully automatic approach for determining the Risser stage using deep learning on abdominal radiographs.

MATERIALS AND METHODS: In this multicenter study, 1,681 supine abdominal radiographs (age range, 9-18 years, 50% female) obtained between January 2019 and April 2022 were collected retrospectively from three medical institutions and graded manually using the United States Risser staging system. A total of 1,577 images from Hospitals 1 and 2 were used for development, and 104 images from Hospital 3 for external validation. From each radiograph, right and left iliac crest patch images were extracted using the pelvic bone segmentation model DeepLabv3 + with the EfficientNet-B0 encoder trained with 90 digitally reconstructed radiographs from pelvic computed tomography scans with a pelvic bone mask. Using these patch images, ConvNeXt-B was trained to grade according to the Risser classification. The model's performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUROC), and mean absolute error.

RESULTS: The fully automatic Risser stage assessment model showed an accuracy of 0.87 and 0.75, mean absolute error of 0.13 and 0.26, and AUROC of 0.99 and 0.95 on internal and external test sets, respectively.

CONCLUSION: We developed a deep learning-based, fully automatic segmentation and classification model for Risser stage assessment using abdominal radiographs.

PMID:39046527 | DOI:10.1007/s00247-024-05999-1

Eur Radiol. 2024 Jul 24. doi: 10.1007/s00330-024-10974-3. Online ahead of print.

ABSTRACT

OBJECTIVES: To perform a multi-reader comparison of multiparametric dual-energy computed tomography (DECT) images reconstructed with deep-learning image reconstruction (DLIR) and standard-of-care adaptive statistical iterative reconstruction-V (ASIR-V).

METHODS: This retrospective study included 100 patients undergoing portal venous phase abdominal CT on a rapid kVp switching DECT scanner. Six reconstructed DECT sets (ASIR-V and DLIR, each at three strengths) were generated. Each DECT set included 65 keV monoenergetic, iodine, and virtual unenhanced (VUE) images. Using a Likert scale, three radiologists performed qualitative assessments for image noise, contrast, small structure visibility, sharpness, artifact, and image preference. Quantitative assessment was performed by measuring attenuation, image noise, and contrast-to-noise ratios (CNR). For the qualitative analysis, Gwet's AC2 estimates were used to assess agreement.

RESULTS: DECT images reconstructed with DLIR yielded better qualitative scores than ASIR-V images except for artifacts, where both groups were comparable. DLIR-H images were rated higher than other reconstructions on all parameters (p-value < 0.05). On quantitative analysis, there was no significant difference in the attenuation values between ASIR-V and DLIR groups. DLIR images had higher CNR values for the liver and portal vein, and lower image noise, compared to ASIR-V images (p-value < 0.05). The subgroup analysis of patients with large body habitus (weight ≥ 90 kg) showed similar results to the study population. Inter-reader agreement was good-to-very good overall.

CONCLUSION: Multiparametric post-processed DECT datasets reconstructed with DLIR were preferred over ASIR-V images with DLIR-H yielding the highest image quality scores.

CLINICAL RELEVANCE STATEMENT: Deep-learning image reconstruction in dual-energy CT demonstrated significant benefits in qualitative and quantitative image metrics compared to adaptive statistical iterative reconstruction-V.

KEY POINTS: Dual-energy CT (DECT) images reconstructed using deep-learning image reconstruction (DLIR) showed superior qualitative scores compared to adaptive statistical iterative reconstruction-V (ASIR-V) reconstructed images, except for artifacts where both reconstructions were rated comparable. While there was no significant difference in attenuation values between ASIR-V and DLIR groups, DLIR images showed higher contrast-to-noise ratios (CNR) for liver and portal vein, and lower image noise (p value < 0.05). Subgroup analysis of patients with large body habitus (weight ≥ 90 kg) yielded similar findings to the overall study population.

PMID:39046499 | DOI:10.1007/s00330-024-10974-3

Radiol Artif Intell. 2024 Jul 24:e230433. doi: 10.1148/ryai.230433. Online ahead of print.

ABSTRACT

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To assess the prognostic value of a deep learning-based chest radiographic age (CXR-Age) model in a large external test cohort of Asian individuals. Materials and Methods This single-center, retrospective study included chest radiographs from consecutive, asymptomatic Asian individuals aged 50 to 80 who received health check-ups between January 2004 and June 2018. This study performed a dedicated external test of the previously developed CXR-Age model, which predicts an age adjusted based on the risk of all-cause mortality. Adjusted hazard ratios (HRs) of CXR-Age for all-cause, cardiovascular, lung cancer, and respiratory disease mortality were assessed using multivariable Cox or Fine-Gray models, with their added values evaluated by likelihood ratio tests. Results A total of 36,924 individuals (mean chronological age ± SD, 58 ± 7 years; CXR-Age, 60 ± 5 years; 22,352 male) were included. Over a median follow-up of 11.0 years, 1250 individuals (3.4%) died, including 153 cardiovascular (0.4%), 166 lung cancer (0.4%), and 98 respiratory deaths (0.3%). CXR-Age was a significant risk factor for all-cause (adjusted HR at the chronological age of 50 years: 1.03; at 60 years: 1.05; at 70 years: 1.07), cardiovascular (adjusted HR: 1.11), lung cancer (adjusted HR for former smokers: 1.12; for current smokers: 1.05), and respiratory disease mortality (adjusted HR: 1.12) (all P values < 0.05). The likelihood ratio test demonstrated added prognostic value of CXR-Age to clinical factors including chronological age for all outcomes (all P values < 0.001). Conclusion Deep learning-based chest radiographic age was associated with various survival outcomes and had added value to clinical factors in asymptomatic Asian individuals, suggesting its generalizability. ©RSNA, 2024.

PMID:39046324 | DOI:10.1148/ryai.230433

CNS Neurosci Ther. 2024 Jul;30(7):e14841. doi: 10.1111/cns.14841.

ABSTRACT

Cerebral small vessel disease (CSVD) is an important cause of stroke, cognitive impairment, and other diseases, and its early quantitative evaluation can significantly improve patient prognosis. Magnetic resonance imaging (MRI) is an important method to evaluate the occurrence, development, and severity of CSVD. However, the diagnostic process lacks quantitative evaluation criteria and is limited by experience, which may easily lead to missed diagnoses and misdiagnoses. With the development of artificial intelligence technology based on deep learning, the extraction of high-dimensional features in imaging can assist doctors in clinical decision-making, and it has been widely used in brain function and mental disorders, and cardiovascular and cerebrovascular diseases. This paper summarizes the global research results in recent years and briefly describes the application of deep learning in evaluating CSVD signs in MRI imaging, including recent small subcortical infarcts, lacunes of presumed vascular origin, vascular white matter hyperintensity, enlarged perivascular spaces, cerebral microbleeds, brain atrophy, cortical superficial siderosis, and cortical cerebral microinfarct.

PMID:39045778 | DOI:10.1111/cns.14841

Microsyst Nanoeng. 2024 Jul 22;10:103. doi: 10.1038/s41378-024-00739-9. eCollection 2024.

ABSTRACT

Wearable devices, such as data gloves and electronic skins, can perceive human instructions, behaviors and even emotions by tracking a hand's motion, with the help of knowledge learning. The shape or position single-mode sensor in such devices often lacks comprehensive information to perceive interactive gestures. Meanwhile, the limited computing power of wearable applications restricts the multimode fusion of different sensing data and the deployment of deep learning networks. We propose a perceptive fusion electronic skin (PFES) with a bioinspired hierarchical structure that utilizes the magnetization state of a magnetostrictive alloy film to be sensitive to external strain or magnetic field. Installed at the joints of a hand, the PFES realizes perception of curvature (joint shape) and magnetism (joint position) information by mapping corresponding signals to the two-directional continuous distribution such that the two edges represent the contributions of curvature radius and magnetic field, respectively. By autonomously selecting knowledge closer to the user's hand movement characteristics, the reinforced knowledge distillation method is developed to learn and compress a teacher model for rapid deployment on wearable devices. The PFES integrating the autonomous learning algorithm can fuse curvature-magnetism dual information, ultimately achieving human machine interaction with gesture recognition and haptic feedback for cross-space perception and manipulation.

PMID:39045231 | PMC:PMC11263581 | DOI:10.1038/s41378-024-00739-9

Front Physiol. 2024 Jul 9;15:1380459. doi: 10.3389/fphys.2024.1380459. eCollection 2024.

ABSTRACT

Introduction: Alzheimer's Disease (AD) is a degenerative brain disorder characterized by cognitive and memory dysfunctions. The early detection of AD is necessary to reduce the mortality rate through slowing down its progression. The prevention and detection of AD is the emerging research topic for many researchers. The structural Magnetic Resonance Imaging (sMRI) is an extensively used imaging technique in detection of AD, because it efficiently reflects the brain variations. Methods: Machine learning and deep learning models are widely applied on sMRI images for AD detection to accelerate the diagnosis process and to assist clinicians for timely treatment. In this article, an effective automated framework is implemented for early detection of AD. At first, the Region of Interest (RoI) is segmented from the acquired sMRI images by employing Otsu thresholding method with Tunicate Swarm Algorithm (TSA). The TSA finds the optimal segmentation threshold value for Otsu thresholding method. Then, the vectors are extracted from the RoI by applying Local Binary Pattern (LBP) and Local Directional Pattern variance (LDPv) descriptors. At last, the extracted vectors are passed to Deep Belief Networks (DBN) for image classification. Results and Discussion: The proposed framework achieves supreme classification accuracy of 99.80% and 99.92% on the Alzheimer's Disease Neuroimaging Initiative (ADNI) and Australian Imaging, Biomarker and Lifestyle flagship work of ageing (AIBL) datasets, which is higher than the conventional detection models.

PMID:39045216 | PMC:PMC11263168 | DOI:10.3389/fphys.2024.1380459

BME Front. 2024 Jul 23;5:0048. doi: 10.34133/bmef.0048. eCollection 2024.

ABSTRACT

Objective and Impact Statement: Human epidermal growth factor receptor 2 (HER2) is a critical protein in cancer cell growth that signifies the aggressiveness of breast cancer (BC) and helps predict its prognosis. Here, we introduce a deep learning-based approach utilizing pyramid sampling for the automated classification of HER2 status in immunohistochemically (IHC) stained BC tissue images. Introduction: Accurate assessment of IHC-stained tissue slides for HER2 expression levels is essential for both treatment guidance and understanding of cancer mechanisms. Nevertheless, the traditional workflow of manual examination by board-certified pathologists encounters challenges, including inter- and intra-observer inconsistency and extended turnaround times. Methods: Our deep learning-based method analyzes morphological features at various spatial scales, efficiently managing the computational load and facilitating a detailed examination of cellular and larger-scale tissue-level details. Results: This approach addresses the tissue heterogeneity of HER2 expression by providing a comprehensive view, leading to a blind testing classification accuracy of 84.70%, on a dataset of 523 core images from tissue microarrays. Conclusion: This automated system, proving reliable as an adjunct pathology tool, has the potential to enhance diagnostic precision and evaluation speed, and might substantially impact cancer treatment planning.

PMID:39045139 | PMC:PMC11265840 | DOI:10.34133/bmef.0048

iScience. 2024 Jun 18;27(7):110279. doi: 10.1016/j.isci.2024.110279. eCollection 2024 Jul 19.

ABSTRACT

Breast cancer is the second leading cause of carcinoma-linked death in women. We developed a multi-modal deep-learning model (BreNet) to differentiate breast cancer from benign lesions. BreNet was constructed and trained on 10,108 images from one center and tested on 3,762 images from two centers in three steps. The diagnostic ability of BreNet was first compared with that of six radiologists; a BreNet-aided scheme was constructed to improve the diagnostic ability of the radiologists; and the diagnosis of real-world radiologists' scheme was then compared with the BreNet-aided scheme. The diagnostic performance of BreNet was superior to that of the radiologists (area under the curve [AUC]: 0.996 vs. 0.841). BreNet-aided scheme increased the pooled AUC of the radiologists from 0.841 to 0.934 for reviewing images, and from 0.892 to 0.934 in the real-world test. The use of BreNet significantly enhances the diagnostic ability of radiologists in the detection of breast cancer.

PMID:39045104 | PMC:PMC11263717 | DOI:10.1016/j.isci.2024.110279

Eur Heart J Imaging Methods Pract. 2023 Nov 27;1(2):qyad040. doi: 10.1093/ehjimp/qyad040. eCollection 2023 Sep.

ABSTRACT

AIMS: Impaired standardization of echocardiograms may increase inter-operator variability. This study aimed to determine whether the real-time guidance of experienced sonographers by deep learning (DL) could improve the standardization of apical recordings.

METHODS AND RESULTS: Patients (n = 88) in sinus rhythm referred for echocardiography were included. All participants underwent three examinations, whereof two were performed by sonographers and the third by cardiologists. In the first study period (Period 1), the sonographers were instructed to provide echocardiograms for the analyses of the left ventricular function. Subsequently, after brief training, the DL guidance was used in Period 2 by the sonographer performing the second examination. View standardization was quantified retrospectively by a human expert as the primary endpoint and the DL algorithm as the secondary endpoint. All recordings were scored in rotation and tilt both separately and combined and were categorized as standardized or non-standardized. Sonographers using DL guidance had more standardized acquisitions for the combination of rotation and tilt than sonographers without guidance in both periods (all P ≤ 0.05) when evaluated by the human expert and DL [except for the apical two-chamber (A2C) view by DL evaluation]. When rotation and tilt were analysed individually, A2C and apical long-axis rotation and A2C tilt were significantly improved, and the others were numerically improved when evaluated by the echocardiography expert. Furthermore, all, except for A2C rotation, were significantly improved when evaluated by DL (P < 0.01).

CONCLUSION: Real-time guidance by DL improved the standardization of echocardiographic acquisitions by experienced sonographers. Future studies should evaluate the impact with respect to variability of measurements and when used by less-experienced operators.

CLINICALTRIALSGOV IDENTIFIER: NCT04580095.

PMID:39045079 | PMC:PMC11195719 | DOI:10.1093/ehjimp/qyad040

Eur Heart J Imaging Methods Pract. 2023 Dec 8;1(2):qyad043. doi: 10.1093/ehjimp/qyad043. eCollection 2023 Sep.

NO ABSTRACT

PMID:39045069 | PMC:PMC11195721 | DOI:10.1093/ehjimp/qyad043

PNAS Nexus. 2024 Jul 23;3(7):pgae243. doi: 10.1093/pnasnexus/pgae243. eCollection 2024 Jul.

ABSTRACT

Volatile organic compounds (VOCs) are ubiquitous in vehicle cabin environments, which can significantly impact the health of drivers and passengers, whereas quick and intelligent prediction methods are lacking. In this study, we firstly analyzed the variations of environmental parameters, VOC levels and potential sources inside a new car during 7 summer workdays, indicating that formaldehyde had the highest concentration and about one third of the measurements exceeded the standard limit for in-cabin air quality. Feature importance analysis reveals that the most important factor affecting in-cabin VOC emission behaviors is the material surface temperature rather than the air temperature. By introducing the attention mechanism and ensemble strategy, we present an LSTM-A-E deep learning model to predict the concentrations of 12 observed typical VOCs, together with other five deep learning models for comparison. By comparing the prediction-observation discrepancies and five evaluation metrics, the LSTM-A-E model demonstrates better performance, which is more consistent with field measurements. Extension of the developed model for predicting the 10-day VOC concentrations in a realistic residence further illustrates its excellent environmental adaptation. This study probes the not-well-explored in-cabin VOC dynamics via observation and deep learning approaches, facilitating rapid prediction and exposure assessment of VOCs in the vehicle micro-environment.

PMID:39045013 | PMC:PMC11264407 | DOI:10.1093/pnasnexus/pgae243

Heliyon. 2024 Jun 24;10(13):e33555. doi: 10.1016/j.heliyon.2024.e33555. eCollection 2024 Jul 15.

ABSTRACT

Aiming at the problems that the traditional image recognition technology is challenging to extract useful features and the recognition time is extended; the AlexNet model is improved to improve the effect of image classification and recognition. This study focuses on 8 types of tomato leaf diseases and healthy leaves. By using HOG and LBP weighted fusion to extract image features, a tomato leaf disease recognition model based on the AlexNet model is proposed, and transfer learning is used to train the AlexNet model. Transfer the knowledge learned by the AlexNet model on the PlantVillage image dataset to this model while reducing the number of fully connected layers. Keras deep learning framework and programming language Python were used. The model was implemented, and the classification and identification of tomato leaf diseases were carried out. The recognition rate of feature-weighted fusion classification is higher than that of serial and parallel methods, and the recognition time is the shortest. When the weight coefficient ratio of HOG and LBP is 3:7, the image recognition rate is the highest, and its value is 97.2 %. From the model performance curve See, when the number of iterations is more than 150 times, the training set and test accuracy rate both exceed 97 %, the loss rate shows a gradient decline, and the change is relatively stable; compared with the traditional AlexNet model, HOG + LBP + SVM model, and VGG model, improved AlexNet model has the highest recognition rate, and it has high recall value, accuracy, and F1 value; Compared with the latest convolutional neural network disease recognition models, improved AlexNet model recognition accuracy was 98.83 %, and the F1 value was 0.994. It shows that the model has good convergence performance, fast prediction speed, and low loss rate and can effectively identify 8 types of tomato leaf images, which provides a reference for the research on crop disease identification.

PMID:39044970 | PMC:PMC11263660 | DOI:10.1016/j.heliyon.2024.e33555

Data Brief. 2024 May 29;55:110564. doi: 10.1016/j.dib.2024.110564. eCollection 2024 Aug.

ABSTRACT

Seasonal vegetables play a crucial role in both nutrition and commerce in Bangladesh. Recognizing this significance, our research introduces the 'SeasVeg' dataset, comprising images of ten varieties of seasonal vegetables sourced from Dhaka and Pabna regions. These include Carica papaya, Momordica dioica, Abelmoschus esculentus, Lablab purpureus, Trichosanthes cucumerina, Trichosanthes dioica, Solanum lycopersicum, Brassica oleracea, Momordica charantia, and Raphanus sativus. Our dataset encompasses 4500 images, 1500 original and 3000 augmented, meticulously captured under natural light conditions to ensure authenticity. While our primary focus lies in leveraging machine learning and deep learning techniques for advancements in agriculture science, particularly in aiding healthcare aspects with seasonal vegetables and nutrition's, we acknowledge the versatile utility of our dataset. Beyond healthcare, it serves as a valuable educational resource, facilitating children's and toddlers' learning to identify these vital vegetables. This dual functionality broadens the dataset's appeal and underscores its societal impact beyond the realm of healthcare. Besides, the research culminates in the implementation of machine learning models, achieving noteworthy accuracy. We get the highest 99 % accuracy with the ResNet50 pre-trained CNN model and a good 94 % accuracy with the InceptionV3 pre-trained CNN model when it comes to the computer-aided vegetable classification. However, the 'SeasVeg' dataset represents not only a significant stride in healthcare innovation but also a promising tool for educational endeavors, catering to diverse stakeholders and fostering interdisciplinary collaboration.

PMID:39044911 | PMC:PMC11263961 | DOI:10.1016/j.dib.2024.110564