Eur Radiol. 2025 Feb 4. doi: 10.1007/s00330-025-11386-7. Online ahead of print.

ABSTRACT

OBJECTIVES: Thyroid cancer, the only cancer that uses age as a specific predictor of survival, is increasing in incidence, yet it has a low mortality rate, which can lead to overdiagnosis and overtreatment. We developed an age-stratified deep learning (DL) model (hereafter, ASMCNet) for classifying thyroid nodules and aimed to investigate the effect of age stratification on the accuracy of a DL model, exploring how ASMCNet can help radiologists improve diagnostic performance and avoid unnecessary biopsies.

METHODS: In this retrospective study, we used ultrasound images from three hospitals, a total of 10,391 images of 5934 patients were used for training, validation, and testing. The performance of ASMCNet was compared with that of model-trained non-age-stratified radiologists with different experience levels on the test data set with the DeLong method.

RESULTS: The area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity of ASMCNet were 0.906, 86.1%, and 85.1%, respectively, which exceeded those of model-trained non-age-stratified (0.867, 83.2%, and 75.5%, respectively; p < 0.001) and higher than all of the radiologists (p < 0.001). Reader studies show that radiologists' performances are improved when assisted by the explaining heatmaps (p < 0.001).

CONCLUSIONS: Our study demonstrates that age stratification based on DL can further improve the performance of thyroid tumor classification models, which also suggests that age is an important factor in the diagnosis of thyroid tumors. The ASMCNet model shows promising clinical applicability and can assist radiologists in improving diagnostic accuracy.

KEY POINTS: Question Age is crucial for differentiated thyroid carcinoma (DTC) prognosis, yet its diagnostic impact lacks research. Findings Adding age stratification to DL models can further improve the accuracy of thyroid nodule diagnosis. Clinical relevance Age-stratified multimodal classification network is a reliable tool used to help radiologists diagnose thyroid nodules, and integrating it into clinical practice can improve diagnostic accuracy and reduce unnecessary biopsies or treatments.

PMID:39903238 | DOI:10.1007/s00330-025-11386-7

Invest Ophthalmol Vis Sci. 2025 Feb 3;66(2):6. doi: 10.1167/iovs.66.2.6.

ABSTRACT

PURPOSE: The purpose of this study was to evaluate the feasibility and utility of optical coherence tomography (OCT)-based, targeted microperimetry grids in assessing focal lesions in intermediate age-related macular degeneration (iAMD).

METHODS: The multicenter, prospective PINNACLE study enrolled 395 patients with iAMD aged 55 to 90 years across 12 international sites. Participants underwent imaging, including OCT and microperimetry, every 4 to 12 months over 3 years. Deep learning algorithms detected focal lesions and changes in OCT images, including drusen regression, EZ/IZ loss with hypertransmission, and subretinal fluid, guiding 5-point microperimetry targeted to lesion locations. Data were analyzed using linear mixed models to estimate differences between retinal sensitivity measured by the 5-point focal grids and sensitivity interpolated from the 24-point standard grids.

RESULTS: The final analysis included 93 eyes from 83 patients, assessing 605 of the 5-point targeted grids and standard grids across 235 focal lesions. The Pearson correlation between focally measured sensitivity and interpolated sensitivity was 0.76. However, interpolation from the standard grid could be erroneous, especially in central regions of lesions characterized by EZ/IZ loss with hypertransmission and subretinal fluid. Interpolation errors increased with distance to the nearest measurement point (slope = 2.20 dB per degree, 95% confidence interval [CI] = 1.52 to 2.87). A significant negative relationship was found between interpolation errors and retinal sensitivity, with the highest errors in areas of low sensitivity. Lesion size significantly impacted interpolation errors for EZ/IZ loss with hypertransmission (slope = -19.41 dB/mm², 95% CI = -29.63 to -9.18).

CONCLUSIONS: Targeted grids improved the detection and understanding of how focal retinal changes affect visual function in patients with iAMD, supporting the development of therapeutic interventions.

PMID:39903180 | DOI:10.1167/iovs.66.2.6

IEEE Access. 2024;12:180792-180814. doi: 10.1109/access.2024.3505038. Epub 2024 Nov 22.

ABSTRACT

Cognitive impairment and dementia-related diseases develop several years before moderate or severe deterioration in cognitive function occurs. Nevertheless, most dementia cases, especially in low- and middle-income countries, remain undiagnosed because of limited access to affordable diagnostic tools. Additionally, the development of accessible tools for diagnosing and predicting cognitive impairment has not been extensively discussed in the literature. The objective of this study is to develop a cost-effective and highly accessible machine learning model to predict the risk of cognitive impairment for up to five years before clinical insight. We utilized easily accessible data from the National Alzheimer's Coordinating Center (NACC) Uniform Data Set (UDS) to train and evaluate various machine learning and deep learning models. A novel algorithm was developed to facilitate the selection of cost-effective models that offer high performance while minimizing development and operational costs. We conducted various assessments, including feature selection, time-series analyses, and external validation of the selected model. Our findings indicated that the Support Vector Machine (SVM) model was preferred over other high-performing neural network models because of its computational efficiency, achieving F2-scores of 0.828 in cross-validation and 0.750 in a generalizability test. Additionally, we found that demographic and historical health data are valuable for early prediction of cognitive impairment. This study demonstrates the potential of developing accessible solutions to predict cognitive impairment early using accurate and efficient machine learning models. Future interventions should consider creating cost-effective assessment tools to support global action plans and reduce the risk of cognitive impairment.

PMID:39902153 | PMC:PMC11790289 | DOI:10.1109/access.2024.3505038

Proc Inst Mech Eng H. 2025 Feb 4:9544119251315877. doi: 10.1177/09544119251315877. Online ahead of print.

ABSTRACT

Generative deep learning has emerged as a promising data augmentation technique in recent years. This approach becomes particularly valuable in areas such as motion analysis, where it is challenging to collect substantial amounts of data. The objective of the current study is to introduce a data augmentation strategy that relies on a variational autoencoder to generate synthetic data of kinetic and kinematic variables. The kinematic and kinetic variables consist of hip and knee joint angles and moments, respectively, in both sagittal and frontal plane, and ground reaction forces. Statistical parametric mapping (SPM) did not detect significant differences between real and synthetic data for each of the biomechanical variables considered. To further evaluate the effectiveness of this approach, a long-short term model (LSTM) was trained both only on real data (R) and on the combination of real and synthetic data (R&S); the performance of each of these two trained models was then assessed on real test data unseen during training. The principal findings included achieving comparable results in terms of nRMSE when predicting knee joint moments in the frontal (R&S: 9.86% vs R: 10.72%) and sagittal plane (R&S: 9.21% vs R: 9.75%), and hip joint moments in the frontal (R&S: 16.93% vs R: 16.79%) and sagittal plane (R&S: 13.29% vs R: 14.60%). The main novelty of this study lies in introducing an effective data augmentation approach in motion analysis settings.

PMID:39902572 | DOI:10.1177/09544119251315877

Anal Chem. 2025 Feb 4. doi: 10.1021/acs.analchem.4c05271. Online ahead of print.

ABSTRACT

Lifetime determination plays a crucial role in fluorescence lifetime imaging microscopy (FLIM). We introduce UNET-FLIM, a deep learning architecture based on a one-dimensional U-net, specifically designed for lifetime determination. UNET-FLIM focuses on handling low photon count data with high background noise levels, which are commonly encountered in fast FLIM applications. The proposed network can be effectively trained using simulated decay curves, making it adaptable to various time-domain FLIM systems. Our evaluations of simulated data demonstrate that UNET-FLIM robustly estimates lifetimes and proportions, even when the signal photon count is extremely low and background noise levels are high. Remarkably, UNET-FLIM's insensitivity to noise and minimal photon count requirements facilitate fast FLIM imaging and real-time lifetime analysis. We demonstrate its potential by applying it to monitor rapid lysosomal pH variations in living cells during in situ acetic acid treatment, all without necessitating any modifications to existing FLIM systems.

PMID:39902564 | DOI:10.1021/acs.analchem.4c05271

Recent Pat Anticancer Drug Discov. 2025 Feb 3. doi: 10.2174/0115748928361472250123105507. Online ahead of print.

ABSTRACT

An aberrant increase in cancer incidences has demanded extreme attention globally despite advancements in diagnostic and management strategies. The high mortality rate is concerning, and tumour heterogeneity at the genetic, phenotypic, and pathological levels exacerbates the problem. In this context, lack of early diagnostic techniques and therapeutic resistance to drugs, sole awareness among the public, coupled with the unavailability of these modern technologies in developing and low-income countries, negatively impact cancer management. One of the prime necessities of the world today is the enhancement of early detection of cancers. Several independent studies have shown that screening individuals for cancer can improve patient survival but are bogged down by risk classification and major problems in patient selection. Artificial intelligence (AI) has significantly advanced the field of oncology, addressing various medical challenges, particularly in cancer management. Leveraging extensive medical datasets and innovative computational technologies, AI, especially through deep learning (DL), has found applications across multiple facets of oncology research. These applications range from early cancer detection, diagnosis, classification, and grading, molecular characterization of tumours, prediction of patient outcomes and treatment responses, personalized treatment, and novel anti-cancer drug discovery. Over the past decade, AI/ML has emerged as a valuable tool in cancer prognosis, risk assessment, and treatment selection for cancer patients. Several patents have been and are being filed and granted. Some of those inventions were explored and are being explored in clinical settings as well. In this review, we will discuss the current status, recent advancements, clinical trials, challenges, and opportunities associated with AI/ML applications in cancer detection and management. We are optimistic about the potential of AI/ML in improving outcomes for cancer and the need for further research and development in this field.

PMID:39902536 | DOI:10.2174/0115748928361472250123105507

J Bone Oncol. 2024 Dec 31;51:100659. doi: 10.1016/j.jbo.2024.100659. eCollection 2025 Apr.

ABSTRACT

BACKGROUND: Colorectal cancer is a prevalent malignancy with a significant risk of metastasis, including to bones, which severely impacts patient outcomes. Accurate prediction of bone metastasis risk is crucial for optimizing treatment strategies and improving prognosis.

PURPOSE: This study aims to develop a predictive model combining radiomics and Vision Transformer (ViT) deep learning techniques to assess the risk of bone metastasis in colorectal cancer patients using both plain and contrast-enhanced CT images.

MATERIALS AND METHODS: We conducted a retrospective analysis of 155 colorectal cancer patients, including 81 with bone metastasis and 74 without. Radiomic features were extracted from segmented tumors on both plain and contrast-enhanced CT images. LASSO regression was applied to select key features, which were then used to build traditional machine learning models, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest, LightGBM, and XGBoost. Additionally, a dual-modality ViT model was trained on the same CT images, with a late fusion strategy employed to combine outputs from the different modalities. Model performance was evaluated using AUC-ROC, accuracy, sensitivity, and specificity, and differences were statistically assessed using DeLong's test.

RESULTS: The ViT model demonstrated superior predictive performance, achieving an AUC of 0.918 on the test set, significantly outperforming all traditional radiomics-based models. The SVM model, while the best among traditional models, still underperformed compared to the ViT model. The ViT model's strength lies in its ability to capture complex spatial relationships and long-range dependencies within the imaging data, which are often missed by traditional models. DeLong's test confirmed the statistical significance of the ViT model's enhanced performance, highlighting its potential as a powerful tool for predicting bone metastasis risk in colorectal cancer patients.

CONCLUSION: The integration of radiomics with ViT-based deep learning offers a robust and accurate method for predicting bone metastasis risk in colorectal cancer patients. The ViT model's ability to analyze dual-modality CT imaging data provides greater precision in risk assessment, which can improve clinical decision-making and personalized treatment strategies. These findings underscore the promise of advanced deep learning models in enhancing the accuracy of metastasis prediction. Further validation in larger, multicenter studies is recommended to confirm the generalizability of these results.

PMID:39902382 | PMC:PMC11787686 | DOI:10.1016/j.jbo.2024.100659

Iran J Public Health. 2025 Jan;54(1):133-143. doi: 10.18502/ijph.v54i1.17583.

ABSTRACT

BACKGROUND: This study was conducted to perform deep learning-based body shape cluster analysis using 3D Body Scanner.

METHODS: For this study, 54 variables were measured using 3D Body Scanner on 366 adult men and women at Korea National Sport University in 2022. Transformer learning and dimensionality reduction models were used to perform cluster analysis on the measured data. Mann-Whitney test and Kruskal-Wallis test were applied to compare the principal component differences of new scale characteristics, and all statistical significance levels were set at .05.

RESULTS: First, among the two methods for classifying body types, the transformer algorithm had a higher performance in body type classification. Second, in the classification of body type clusters, two clusters, endomorphic body type and ectomorphic body type, were divided into six clusters, two for cluster 1 and four for cluster 2.

CONCLUSION: The six clusters provide more granular information than previous body type classifications, and we believe that they can be used as basic information for predicting health and disease.

PMID:39902372 | PMC:PMC11787839 | DOI:10.18502/ijph.v54i1.17583

Front Plant Sci. 2025 Jan 20;15:1514832. doi: 10.3389/fpls.2024.1514832. eCollection 2024.

ABSTRACT

Plant disease detection remains a significant challenge, necessitating innovative approaches to enhance detection efficiency and accuracy. This study proposes an improved YOLOv8 model, SerpensGate-YOLOv8, specifically designed for plant disease detection tasks. Key enhancements include the incorporation of Dynamic Snake Convolution (DySnakeConv) into the C2F module, which improves the detection of intricate features in complex structures, and the integration of the SPPELAN module, combining Spatial Pyramid Pooling (SPP) and Efficient Local Aggregation Network (ELAN) for superior feature extraction and fusion. Additionally, an innovative Super Token Attention (STA) mechanism was introduced to strengthen global feature modeling during the early stages of the network. The model leverages the PlantDoc dataset, a highly generalizable dataset containing 2,598 images across 13 plant species and 27 classes (17 diseases and 10 healthy categories). With these improvements, the model achieved a Precision of 0.719. Compared to the original YOLOv8, the mean Average Precision (mAP@0.5) improved by 3.3%, demonstrating significant performance gains. The results indicate that SerpensGate-YOLOv8 is a reliable and efficient solution for plant disease detection in real-world agricultural environments.

PMID:39902212 | PMC:PMC11788276 | DOI:10.3389/fpls.2024.1514832

Front Plant Sci. 2025 Jan 20;15:1502631. doi: 10.3389/fpls.2024.1502631. eCollection 2024.

ABSTRACT

Rice is an important part of the food supply, its different varieties in terms of quality, flavor, nutritional value, and other aspects of the differences, directly affect the subsequent yield and economic benefits. However, traditional rice identification methods are time-consuming, inefficient, and prone to damage. For this reason, this study proposes a deep learning-based method to classify and identify rice with different flavors in a fast and non-destructive way. In this experiment, 19 categories of japonica rice seeds were selected, and a total of 36735 images were finally obtained. The lightweight network High Precision FasterNet (HPFasterNet) proposed in this study combines the Ghost bottleneck and FasterNet_T0 and introduces group convolution to compare the model performance. The results show that HPFasterNet has the highest classification accuracy of 92%, which is 5.22% better than the original model FasterNet_T0, and the number of parameters and computation is significantly reduced compared to the original model, which is more suitable for resource-limited environments. Comparison with three classical models and three lightweight models shows that HPFasterNet exhibits a more comprehensive and integrated performance. Meanwhile, in this study, HPFasterNet was used to test rice with different flavors, and the accuracy reached 98.98%. The experimental results show that the network model proposed in this study can be used to provide auxiliary experiments for rice breeding and can also be applied to consumer and food industries.

PMID:39902203 | PMC:PMC11788896 | DOI:10.3389/fpls.2024.1502631

Biol Methods Protoc. 2024 Nov 20;10(1):bpae086. doi: 10.1093/biomethods/bpae086. eCollection 2025.

ABSTRACT

Brain age, as a correlate of an individual's chronological age obtained from structural and functional neuroimaging data, enables assessing developmental or neurodegenerative pathology relative to the overall population. Accurately inferring brain age from brain magnetic resonance imaging (MRI) data requires imaging methods sensitive to tissue health and sophisticated statistical models to identify the underlying age-related brain changes. Magnetic resonance elastography (MRE) is a specialized MRI technique which has emerged as a reliable, non-invasive method to measure the brain's mechanical properties, such as the viscoelastic shear stiffness and damping ratio. These mechanical properties have been shown to change across the life span, reflect neurodegenerative diseases, and are associated with individual differences in cognitive function. Here, we aim to develop a machine learning framework to accurately predict a healthy individual's chronological age from maps of brain mechanical properties. This framework can later be applied to understand neurostructural deviations from normal in individuals with neurodevelopmental or neurodegenerative conditions. Using 3D convolutional networks as deep learning models and more traditional statistical models, we relate chronological age as a function of multiple modalities of whole-brain measurements: stiffness, damping ratio, and volume. Evaluations on held-out subjects show that combining stiffness and volume in a multimodal approach achieves the most accurate predictions. Interpretation of the different models highlights important regions that are distinct between the modalities. The results demonstrate the complementary value of MRE measurements in brain age models, which, in future studies, could improve model sensitivity to brain integrity differences in individuals with neuropathology.

PMID:39902188 | PMC:PMC11790219 | DOI:10.1093/biomethods/bpae086

Front Oral Health. 2025 Jan 20;5:1424840. doi: 10.3389/froh.2024.1424840. eCollection 2024.

ABSTRACT

BACKGROUND: The styloid process (SP), a bony projection from the temporal bone which can become elongated, resulting in cervical pain, throat discomfort, and headaches. Associated with Eagle syndrome, this elongation can compress nearby nerves and blood vessels, leading to potentially severe complications. Traditional imaging-based methods for classifying various types of elongated styloid processes (ESP) are challenging due to variations in image quality, patient positioning, and anatomical differences, which limit diagnostic accuracy. Recent advancements in artificial intelligence, particularly deep learning, provide more efficient classification of elongated styloid processes.

OBJECTIVE: This study aims to develop an automated classification system for elongated styloid processes using deep learning models and to evaluate the performance of two distinct architectures, EfficientNetB5 and InceptionV3, in classifying elongated styloid processes.

METHODS: This retrospective analysis classified elongated styloid processes using Ortho Pantomograms (OPG) sourced from our oral radiology archives. Styloid process lengths were measured using ImageJ software. A dataset of 330 elongated and 120 normal styloid images was curated for deep learning model training and testing. Pre-processing included median filtering and resizing, with data augmentation applied to improve generalization. EfficientNetB5 and InceptionV3 models, utilized as feature extractors, captured unique styloid characteristics. Model performance was evaluated based on accuracy, precision, recall, and F1-score, with a comparative analysis conducted to identify the most effective model and support advancements in patient care.

RESULTS: The EfficientNetB5 model achieved an accuracy of 97.49%, a precision of 98.00%, a recall of 97.00%, and an F1-score of 97.00%, demonstrating strong overall performance. Additionally, the model achieved an AUC of 0.9825. By comparison, the InceptionV3 model achieved an accuracy of 84.11%, a precision of 85.00%, a recall of 84.00%, and an F1-score of 84.00%, with an AUC of 0.8943. This comparison indicates that EfficientNetB5 outperformed InceptionV3 across all key metrics.

CONCLUSION: In conclusion, our study presents a deep learning-based approach utilizing EfficientNetB5 and InceptionV3 to accurately categorize elongated styloid processes into distinct types based on their morphological characteristics from digital panoramic radiographs. Our results indicate that these models, particularly EfficientNetB5, can enhance diagnostic accuracy and streamline clinical workflows, contributing to improved patient care.

PMID:39902080 | PMC:PMC11788325 | DOI:10.3389/froh.2024.1424840

Clin Transl Radiat Oncol. 2025 Jan 19;51:100925. doi: 10.1016/j.ctro.2025.100925. eCollection 2025 Mar.

ABSTRACT

The critical aspect of successful brachytherapy (BT) is accurate detection of applicator/needle trajectories, which is an ongoing challenge. This study proposes a two-phase deep learning-based method to automate localization of high-dose-rate (HDR) prostate BT catheters through the patient's CT images. The whole process is divided into two phases using two different deep neural networks. First, BT needles segmentation was accomplished through a pix2pix Generative Adversarial Neural network (pix2pix GAN). Second, a Generic Object Tracking Using Regression Networks (GOTURN) was used to predict the needle trajectories. These models were trained and tested on a clinical prostate BT dataset. Among the total 25 patients, 5 patients that consist of 592 slices was dedicated to testing sets, and the rest were used as train/validation set. The total number of needles in these slices of CT images was 8764, of which the employed pix2pix network was able to segment 98.72 % (8652 of total). Dice Similarity Coefficient (DSC) and IoU (Intersection over Union) between the network output and the ground truth were 0.95 and 0.90, respectively. Moreover, the F1-score, recall, and precision results were 0.95, 0.93, and 0.97, respectively. Regarding the location of the shafts, the proposed model has an error of 0.41 mm. The current study proposed a novel methodology to automatically localize and reconstruct the prostate HDR-BT interstitial needles through the 3D CT images. The presented method can be utilized as a computer-aided module in clinical applications to automatically detect and delineate the multi-catheters, potentially enhancing the treatment quality.

PMID:39901943 | PMC:PMC11788795 | DOI:10.1016/j.ctro.2025.100925

Sci Rep. 2025 Feb 3;15(1):4125. doi: 10.1038/s41598-025-85372-w.

ABSTRACT

In the wagyu industry worldwide, high-quality marbling beef is produced by promoting intramuscular fat deposition during cattle fattening stage through dietary vitamin A control. Thus, however, cattle become susceptible to either vitamin A deficiency or excess state, not only influencing cattle performance and beef quality, but also causing health problems. Researchers have been exploring eye photography monitoring methods for cattle blood vitamin A levels based on the relation between vitamin A and retina colour changes. But previous endeavours cannot realise real-time monitoring and their prediction accuracy still need improvement in a practical sense. This study developed a handheld camera system capable of capturing cattle fundus images and predicting vitamin A levels in real time using deep learning. 4000 fundus images from 50 Japanese Black cattle were used to train and test the prediction algorithms, and the model achieved an average 87%, 83%, and 80% accuracy for three levels of vitamin A deficiency classification (particularly 87% for severe level), demonstrating the effectiveness of camera system in vitamin A deficiency prediction, especially for screening and early warning. More importantly, a new method was exemplified to utilise visualisation heatmap for colour-related DNNs tasks, and it was found that chromatic features extracted from LRP heatmap highlighted-ROI could account for 70% accuracy for the prediction of vitamin A deficiency. This system can assist farmers in blood vitamin A level monitoring and related disease prevention, contributing to precision livestock management and animal well-being in wagyu industry.

PMID:39900776 | DOI:10.1038/s41598-025-85372-w

NPJ Precis Oncol. 2025 Feb 3;9(1):35. doi: 10.1038/s41698-025-00804-0.

ABSTRACT

The rapid development of deep learning has revolutionized medical image processing, including analyzing whole slide images (WSIs). Despite the demonstrated potential for characterizing gene mutations directly from WSIs in certain cancers, challenges remain due to image resolution and reliance on manual annotations for acute myeloid leukemia (AML). We, therefore, propose a deep learning model based on multiple instance learning (MIL) with ensemble techniques to predict gene mutations from AML WSIs. Our model predicts NPM1 mutations and FLT3-ITD without requiring patch-level or cell-level annotations. Using a dataset of 572 WSIs, the largest database with both WSI and genetic mutation information, our model achieved an AUC of 0.90 ± 0.08 for NPM1 and 0.80 ± 0.10 for FLT3-ITD in the testing cohort. Additionally, we found that blasts are pivotal indicators for gene mutation predictions, with their proportions varying between mutated and standard WSIs, highlighting the clinical potential of AML WSI analysis.

PMID:39900774 | DOI:10.1038/s41698-025-00804-0

BMC Womens Health. 2025 Feb 3;25(1):47. doi: 10.1186/s12905-025-03569-7.

ABSTRACT

BACKGROUND: Timely diagnosis is a crucial factor in decreasing the death rate of patients with breast cancer. BI-RADS categories IVc and V indicate a strong suspicion of cancer. The categorisation of each group is determined by the characteristics of the lesion. Certain benign breast lesions might have radiological features indicative of malignancy; thus, biopsy is mandatory. This study aimed to identify the histopathological diagnosis of benign breast masses classified into BIRADS IVc and V subgroups, investigate the radiological characteristics of these masses, and identify ultrasound features that could lead to false positive results (benign lesions that mimic malignancy on imaging).

METHODS: This was a retrospective cross-sectional study at a single facility. Breast lesions reported as BIRADS IVc and V that underwent needle core/stereotactic vacuum-assisted biopsy were reviewed. Patients with benign pathologic diagnoses were analysed, delineating pathological diagnoses. Radiological descriptors were compared to those of a matched control of 50 malignant cases with BIRADS IVc.

RESULTS: A total of 828 breast lesions classified as BIRADS IVc or V were detected during the period spanning from 2015 to 2022. Forty-four lesions (44/828, 5.3%) were benign at initial biopsy, while 784 lesions (784/828, 94.7%) were malignant. After histopathological testing and repeat biopsy, 26/828 (3.14%) patients had discordant benign diagnosis. Half of the repeated biopsies (10/20, 50%) showed malignant pathology. Compared to that in the control group, the presence of an oval shape of the mass was significantly more common in patients with benign pathology (p = 0.035). Conversely, the presence of posterior shadowing was significantly less common (p = 0.050) in benign lesions. No significant differences were observed for the other radiological characteristics. The most common histopathological diagnosis was fibrocystic change.

CONCLUSION: This study highlights key findings regarding the sonographic imaging descriptors and histopathological diagnoses of benign breast lesions categorised as BIRADS IVc/V. The study recommends a correlation between clinical and radiological findings and encourages multidisciplinary decision-making among radiologists, pathologists, and clinicians to determine if a repeat biopsy is warranted. There is a need for continuous research to improve the diagnosis and treatment of breast lesions and reduce false-positive rates by incorporating other methodologies such as sonoelastography and incorporating deep learning and artificial intelligence in the decision-making to eliminate unnecessary procedures.

PMID:39901102 | DOI:10.1186/s12905-025-03569-7

BMC Med Imaging. 2025 Feb 3;25(1):37. doi: 10.1186/s12880-025-01578-4.

ABSTRACT

Liver cancer detection is critically important in the discipline of biomedical image testing and diagnosis. Researchers have explored numerous machine learning (ML) techniques and deep learning (DL) approaches aimed at the automated recognition of liver disease by analysing computed tomography (CT) images. This study compares two frameworks, Deep Convolutional Neural Network (DCNN) and Hierarchical Fusion Convolutional Neural Networks (HFCNN), to assess their effectiveness in liver cancer segmentation. The contribution includes enhancing the edges and textures of CT images through filtering to achieve precise liver segmentation. Additionally, an existing DL framework was employed for liver cancer detection and segmentation. The strengths of this paper include a clear emphasis on the criticality of liver cancer detection in biomedical imaging and diagnostics. It also highlights the challenges associated with CT image detection and segmentation and provides a comprehensive summary of recent literature. However, certain difficulties arise during the detection process in CT images due to overlapping structures, such as bile ducts, blood vessels, image noise, textural changes, size and location variations, and inherent heterogeneity. These factors may lead to segmentation errors and subsequently different analyses. This research analysis compares two advanced methodologies, DCNN and HFCNN, for liver cancer detection. The evaluation of DCNN and HFCNN in liver cancer detection is conducted using multiple performance metrics, including precision, F1-score, recall, and accuracy. This comprehensive assessment provides a detailed evaluation of these models' effectiveness compared to other state-of-the-art methods in identifying liver cancer.

PMID:39901085 | DOI:10.1186/s12880-025-01578-4

Sci Rep. 2025 Feb 3;15(1):4080. doi: 10.1038/s41598-025-88492-5.

ABSTRACT

Epilepsy is a common neurological disorder in which abnormal brain waves propagate rapidly in the brain in the form of a graph network during seizures, and seizures are extremely sudden. So, designing accurate and reliable prediction methods can provide early warning for patients, which is crucial for improving their lives. In recent years, a large number of studies have been conducted using deep learning models on epileptic open electroencephalogram (EEG) datasets with good results, but due to individual differences there are still some subjects whose seizure features cannot be accurately captured and are more difficult to differentiate, with poor prediction results. Important time-varying information may be overlooked if only graph space features during seizures are considered. To address these issues, we propose a synchronization-based graph spatio-temporal attention network (SGSTAN). This model effectively leverages the intricate information embedded within EEG recordings through spatio-temporal correlations. Experimental results on public datasets demonstrate the efficacy of our approach. On the CHB-MIT dataset, our method achieves accuracy, specificity, and sensitivity scores of 98.2%, 98.07%, and 97.85%, respectively. In the case of challenging subjects that are difficult to classify, we achieved an outstanding average classification accuracy of 97.59%, surpassing the results of previous studies.

PMID:39901056 | DOI:10.1038/s41598-025-88492-5

Sci Rep. 2025 Feb 3;15(1):4058. doi: 10.1038/s41598-025-87824-9.

ABSTRACT

With the rapid advancement of artificial intelligence, questions are increasingly being raised by stakeholders regarding how such technologies can enhance the environmental, social, and governance outcomes of organizations. In this study, challenges related to the organization and retrieval of video content within large, heterogeneous media archives are addressed. Existing methods, often reliant on human intervention or low-complexity algorithms, are observed to struggle with the growing demands of online video quantity and quality. To address these limitations, a novel approach is proposed, where convolutional neural networks and long short-term memory networks are utilized to extract both frame-level and temporal video features. Residual networks 50 (ResNet50) is integrated for enhanced content representation, and two-frame video flow is employed to improve system performance. The framework achieves precision, recall, and F-score of 79.2%, 86.5%, and 83%, respectively, on the YouTube, EPFL, and TVSum datasets. Beyond technological advancements, opportunities for effective content management are highlighted, emphasizing the promotion of sustainable digital practices. By minimizing data duplication and optimizing resource usage, scalable solutions for large media collections are supported by the proposed system.

PMID:39901035 | DOI:10.1038/s41598-025-87824-9

Sci Rep. 2025 Feb 3;15(1):4089. doi: 10.1038/s41598-025-88446-x.

ABSTRACT

One of the most significant, intricate, and little-discussed aspects of pressurized irrigation is cost estimation. This study attempts to model the early-stage cost of the drip irrigation system using a database of 515 projects divided into four sections the cost of the pumping station and central control system (TCP), the cost of on-farm equipment (TCF), the cost of installation and operation on-farm and pumping station (TCI), and the total cost (TCT). First, 39 environmental and management features affecting the cost of the listed sectors were extracted for each of the 515 projects previously mentioned. A database (a matrix of 515 × 43) was created, and the costs of all projects were updated for the baseline year of 2022. Then, several feature selection algorithms, such as WCC, LCA, GA, PSO, ACO, ICA, LA, HTS, FOA, DSOS, and CUK, were employed to choose the most significant features that had the biggest influence on the system cost. The selection of features was carried out for all features (a total of 39 features) as well as for easily available features (those features that existed before the irrigation system's design phase, 18 features). Then, different machine learning models such as Multivariate Linear Regression, Support Vector Regression, Artificial Neural Networks, Gene Expression Programming, Genetic Algorithms, Deep Learning, and Decision Trees, were used to estimate the costs of each of the of the aforementioned sections. Support vector machine (SVM) and optimization algorithms (Wrapper) were found to be the best learner and feature selection techniques, respectively, out of all the available feature selection algorithms. The two LCA and FOA algorithms produced the best estimation, according to the evaluation criteria results. Their RMSE for all features was 0.0020 and 0.0018, respectively, and their R2 was 0.94 and 0.94. For readily available features, these criteria were 0.0006 and 0.95 for both algorithms. In the part of the overall feature, the early-stage cost modeling with selected features revealed that the SVM model (with RBF Kernel) is the best model among the four cost sections discussed. Its evaluation criteria in the training stage are R2 = 0.923, RMSE = 0.008, and VE = 0.082; in the testing stage, they are R2 = 0.893, RMSE = 0.009, and VE = 0.102. The ANN model (MLP) was found to be the best model for a subset of features in the easily available feature part, with R2 = 0.912, RMSE = 0.008, and VE = 0.083 in the training stage and R2 = 0.882, RMSE = 0.009, and VE = 0.103 in the testing stage. The findings of this study can be utilized to highly accurately estimate the cost of local irrigation systems based on the recognized environmental and management parameters and by employing particular models.

PMID:39900997 | DOI:10.1038/s41598-025-88446-x