Comput Biol Chem. 2024 Oct 28;113:108270. doi: 10.1016/j.compbiolchem.2024.108270. Online ahead of print.

ABSTRACT

The oral route is the most preferred route for drug delivery, due to which the largest share of the pharmaceutical market is represented by oral drugs. Human intestinal absorption (HIA) is closely related to oral bioavailability making it an important factor in predicting drug absorption. In this study, we focus on predicting drug permeability at HIA as a marker for oral bioavailability. A set of 2648 compounds were collected from some early as well as recent works and curated to build a robust dataset. Five machine learning (ML) algorithms have been trained with a set of molecular descriptors of these compounds which have been selected after rigorous feature engineering. Additionally, two deep learning models - graph convolution neural network (GCNN) and graph attention network (GAT) based model were developed using the same set of compounds to exploit the predictability with automated extracted features. The numerical analyses show that out the five ML models, Random forest and LightGBM could predict with an accuracy of 87.71 % and 86.04 % on the test set and 81.43 % and 77.30 % with the external validation set respectively. Whereas with the GCNN and GAT based models, the final accuracy achieved was 77.69 % and 78.58 % on test set and 79.29 % and 79.42 % on the external validation set respectively. We believe deployment of these models for screening oral drugs can provide promising results and therefore deposited the dataset and models on the GitHub platform (https://github.com/hridoy69/HIA).

PMID:39481232 | DOI:10.1016/j.compbiolchem.2024.108270

Radiography (Lond). 2024 Oct 30;30 Suppl 2:79-87. doi: 10.1016/j.radi.2024.10.010. Online ahead of print.

ABSTRACT

INTRODUCTION: In Autumn 2023, amendments to the Health and Care Professions Councils (HCPC) Standards of Proficiency for Radiographers were introduced requiring clinicians to demonstrate awareness of the principles of AI and deep learning technology, and its application to practice' (HCPC 2023; standard 12.25). With the rapid deployment of AI in departments, staff must be prepared to implement and utilise AI. AI readiness is crucial for adoption, with education as a key factor in overcoming fear and resistance. This survey aimed to assess the current understanding of AI among students and qualified staff in clinical practice.

METHODS: A survey targeting radiographers (diagnostic and therapeutic), radiologists and students was conducted to gather demographic data and assess awareness of AI in clinical practice. Hosted online via JISC, the survey included both closed and open-ended questions and was launched in March 2023 at the European Congress of Radiology (ECR).

RESULTS: A total of 136 responses were collected from participants across 25 countries and 5 continents. The majority were diagnostic radiographers 56.6 %, followed by students 27.2 %, dual-qualified 3.7 % and radiologists 2.9 %. Of the respondents, 30.1 % of respondents indicated that their highest level of qualification was a Bachelor's degree, 29.4 % stated that they are currently using AI in their role, whilst 27 % were unsure. Only 10.3 % had received formal AI training.

CONCLUSION: This study reveals significant gaps in training and understanding of AI among medical imaging staff. These findings will guide further research into AI education for medical imaging professionals.

IMPLICATIONS FOR PRACTICE: This paper lays foundations for future qualitative studies on the provision of AI education for medical imaging professionals, helping to prepare the workforce for the evolving role of AI in medical imaging.

PMID:39481214 | DOI:10.1016/j.radi.2024.10.010

Neural Netw. 2024 Oct 18;181:106809. doi: 10.1016/j.neunet.2024.106809. Online ahead of print.

ABSTRACT

Single underwater images often face limitations in field-of-view and visual perception due to scattering and absorption. Numerous image stitching techniques have attempted to provide a wider viewing range, but the resulting stitched images may exhibit unsightly irregular boundaries. Unlike natural landscapes, the absence of reliable high-fidelity references in water complicates the replicability of these deep learning-based methods, leading to unpredictable distortions in cross-domain applications. To address these challenges, we propose an Underwater Wide-field Image Rectangling and Enhancement (UWIRE) framework that incorporates two procedures, i.e., the R-procedure and E-procedure, both of which employ self-coordinated modes, requiring only a single underwater stitched image as input. The R-procedure rectangles the irregular boundaries in stitched images by employing the initial shape resizing and mesh-based image preservation warping. Instead of local linear constraints, we use complementary optimization of boundary-structure-content to ensure a natural appearance with minimal distortion. The E-procedure enhances the rectangled image by employing parameter-adaptive correction to balance information distribution across channels. We further propose an attentive weight-guided fusion method to balance the perception of color restoration, contrast enhancement, and texture sharpening in a complementary manner. Comprehensive experiments demonstrate the superior performance of our UWIRE framework over state-of-the-art image rectangling and enhancement methods, both in quantitative and qualitative evaluation.

PMID:39481203 | DOI:10.1016/j.neunet.2024.106809

Science. 2024 Nov;386(6721):eadq4946. doi: 10.1126/science.adq4946. Epub 2024 Nov 1.

ABSTRACT

The AlphaFold Protein Structure Database (AFDB) contains more than 214 million predicted protein structures composed of domains, which are independently folding units found in multiple structural and functional contexts. Identifying domains can enable many functional and evolutionary analyses but has remained challenging because of the sheer scale of the data. Using deep learning methods, we have detected and classified every domain in the AFDB, producing The Encyclopedia of Domains. We detected nearly 365 million domains, over 100 million more than can be found by sequence methods, covering more than 1 million taxa. Reassuringly, 77% of the nonredundant domains are similar to known superfamilies, greatly expanding representation of their domain space. We uncovered more than 10,000 new structural interactions between superfamilies and thousands of new folds across the fold space continuum.

PMID:39480926 | DOI:10.1126/science.adq4946

PLoS One. 2024 Oct 31;19(10):e0313011. doi: 10.1371/journal.pone.0313011. eCollection 2024.

ABSTRACT

PURPOSE: This study compared field-of-view (FOV) optimized and constrained undistorted single-shot diffusion-weighted imaging (FOCUS DWI) with deep-learning-based reconstruction (DLR) to conventional DWI for breast imaging.

METHODS: This study prospectively enrolled 49 female patients suspected of breast cancer from July to December 2023. The patients underwent conventional and FOCUS breast DWI and data were reconstructed with and without DLR. Two radiologists independently evaluated three images per patient using a 5-point Likert scale. Objective evaluations, including signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and apparent diffusion coefficient (ADC), were conducted using manual region of interest-based analysis. The subjective and objective evaluations were compared using the Friedman test.

RESULTS: The scores for the overall image quality, anatomical details, lesion conspicuity, artifacts, and distortion in FOCUS-DLR DWI were higher than in conventional DWI (all P < 0.001). The SNR of FOCUS-DLR DWI was higher than that of conventional and FOCUS DWI (both P < 0.001), while FOCUS and conventional DWI were similar (P = 0.096). Conventional, FOCUS, and FOCUS-DLR DWI had similar CNR and ADC values.

CONCLUSION: Our findings indicate that images produced by FOCUS-DLR DWI were superior to conventional DWI, supporting the applicability of this technique in clinical practice. DLR provides a new approach to optimize breast DWI.

PMID:39480865 | DOI:10.1371/journal.pone.0313011

IEEE J Biomed Health Inform. 2024 Oct 31;PP. doi: 10.1109/JBHI.2024.3489626. Online ahead of print.

ABSTRACT

Hepatic steatosis, a key factor in chronic liver diseases, is difficult to diagnose early. This study introduces a classifier for hepatic steatosis using microwave technology, validated through clinical trials. Our method uses microwave signals and deep learning to improve detection to reliable results. It includes a pipeline with simulation data, a new deep-learning model called HepNet, and transfer learning. The simulation data, created with 3D electromagnetic tools, is used for training and evaluating the model. HepNet uses skip connections in convolutional layers and two fully connected layers for better feature extraction and generalization. Calibration and uncertainty assessments ensure the model's robustness. Our simulation achieved an F1-score of 0.91 and a confidence level of 0.97 for classifications with entropy ≤0.1, outperforming traditional models like LeNet (0.81) and ResNet (0.87). We also use transfer learning to adapt HepNet to clinical data with limited patient samples. Using 1H-MRS as the standard for two microwave liver scanners, HepNet achieved high F1-scores of 0.95 and 0.88 for 94 and 158 patient samples, respectively, showing its clinical potential.

PMID:39480722 | DOI:10.1109/JBHI.2024.3489626

IEEE Trans Ultrason Ferroelectr Freq Control. 2024 Oct 31;PP. doi: 10.1109/TUFFC.2024.3488729. Online ahead of print.

ABSTRACT

Ultrasound microvascular imaging (UMI), including ultrafast power Doppler imaging (uPDI) and ultrasound localization microscopy (ULM), obtains blood flow information through plane wave transmissions at high frame rates. However, low signal-to-noise ratio of plane waves causes low image quality. Adaptive beamformers have been proposed to suppress noise energy to achieve higher image quality accompanied by increasing computational complexity. Deep learning (DL) leverages powerful hardware capabilities to enable rapid implementation of noise suppression at the cost of flexibility. To enhance the applicability of DL-based methods, in this work, we propose a deep power-aware tunable (DPT) weighting (i.e., postfilter) for delay-and-sum (DAS) beamforming to improve UMI by enhancing plane wave images. The model, called Yformer is a hybrid structure combining convolution and Transformer. With the DAS beamformed and compounded envelope image as input, Yformer can estimate both noise power and signal power. Furthermore, we utilize the obtained powers to compute pixel-wise weights by introducing a tunable noise control factor, which is tailored for improving the quality of different UMI applications. In vivo experiments on the rat brain demonstrate that Yformer can accurately estimate the powers of noise and signal with the structural similarity index (SSIM) higher than 0.95. The performance of the DPT weighting is comparable to that of superior adaptive beamformer in uPDI with low computational cost. The DPT weighting was then applied to four different datasets of ULM, including public simulation, public rat brain, private rat brain, and private rat liver datasets, showing excellent generalizability using the model trained by the private rat brain dataset only. In particular, our method indirectly improves the resolution of liver ULM from 25.24 μm to 18.77 μm by highlighting small vessels. In addition, the DPT weighting exhibits more details of blood vessels with faster processing, which has the potential to facilitate the clinical applications of high-quality UMI.

PMID:39480714 | DOI:10.1109/TUFFC.2024.3488729

IEEE Trans Neural Netw Learn Syst. 2024 Oct 31;PP. doi: 10.1109/TNNLS.2024.3485529. Online ahead of print.

ABSTRACT

Network complexity and computational efficiency have become increasingly significant aspects of deep learning. Sparse deep learning addresses these challenges by recovering a sparse representation of the underlying target function by reducing heavily overparameterized deep neural networks. Specifically, deep neural architectures compressed via structured sparsity (e.g., node sparsity) provide low-latency inference, higher data throughput, and reduced energy consumption. In this article, we explore two well-established shrinkage techniques, Lasso and Horseshoe, for model compression in Bayesian neural networks (BNNs). To this end, we propose structurally sparse BNNs, which systematically prune excessive nodes with the following: 1) spike-and-slab group Lasso (SS-GL) and 2) SS group Horseshoe (SS-GHS) priors, and develop computationally tractable variational inference, including continuous relaxation of Bernoulli variables. We establish the contraction rates of the variational posterior of our proposed models as a function of the network topology, layerwise node cardinalities, and bounds on the network weights. We empirically demonstrate the competitive performance of our models compared with the baseline models in prediction accuracy, model compression, and inference latency.

PMID:39480710 | DOI:10.1109/TNNLS.2024.3485529

Ann Biomed Eng. 2024 Oct 31. doi: 10.1007/s10439-024-03640-8. Online ahead of print.

ABSTRACT

The Fontan procedure is the definitive palliation for pediatric patients born with single ventricles. Surgical planning for the Fontan procedure has emerged as a promising vehicle toward optimizing outcomes, where pre-operative measurements are used prospectively as post-operative boundary conditions for simulation. Nevertheless, actual post-operative measurements can be very different from pre-operative states, which raises questions for the accuracy of surgical planning. The goal of this study is to apply machine leaning techniques to describing pre-operative and post-operative vena caval flow conditions in Fontan patients in order to develop predictions of post-operative boundary conditions to be used in surgical planning. Based on a virtual cohort synthesized by lumped-parameter models, we proposed a novel diversity-aware generative adversarial active learning framework to successfully train predictive deep neural networks on very limited amount of cases that are generally faced by cardiovascular studies. Results of 14 groups of experiments uniquely combining different data query strategies, metrics, and data augmentation options with generative adversarial networks demonstrated that the highest overall prediction accuracy and coefficient of determination were exhibited by the proposed method. This framework serves as a first step toward deep learning for cardiovascular flow prediction/regression with reduced labeling requirements and augmented learning space.

PMID:39480609 | DOI:10.1007/s10439-024-03640-8

Theranostics. 2024 Oct 14;14(17):6818-6830. doi: 10.7150/thno.100298. eCollection 2024.

ABSTRACT

Background: Dynamic real-time detection of dendritic cell (DC) maturation is pivotal for accurately predicting immune system activation, assessing vaccine efficacy, and determining the effectiveness of immunotherapy. The heterogeneity of cells underscores the significance of assessing the maturation status of each individual cell, while achieving real-time monitoring of DC maturation at the single-cell level poses significant challenges. Surface-enhanced Raman spectroscopy (SERS) holds great potential for providing specific fingerprinting information of DCs to detect biochemical alterations and evaluate their maturation status. Methods: We developed Au@CpG@PEG nanoparticle as a self-reporting nanovaccine for DC activation and maturation state assessment, utilizing a label-free SERS strategy. Fingerprint vibrational spectra of the biological components in different states of DCs were collected and analyzed using deep learning Convolutional Neural Networks (CNN) algorithms, aiding in the rapid and efficient identification of DC maturation. Results: This approach enables dynamic real-time detection of DC maturation, maintaining accuracy levels above 98.92%. Conclusion: By employing molecular profiling, we revealed that the signal ratio of tryptophan-to-carbohydrate holds potential as a prospective marker for distinguishing the maturation status of DCs.

PMID:39479453 | PMC:PMC11519801 | DOI:10.7150/thno.100298

Front Digit Health. 2024 Oct 16;6:1455767. doi: 10.3389/fdgth.2024.1455767. eCollection 2024.

ABSTRACT

INTRODUCTION: Extraction of Doppler-based measurements from feto-placental Doppler images is crucial in identifying vulnerable new-borns prenatally. However, this process is time-consuming, operator dependent, and prone to errors.

METHODS: To address this, our study introduces an artificial intelligence (AI) enabled workflow for automating feto-placental Doppler measurements from four sites (i.e., Umbilical Artery (UA), Middle Cerebral Artery (MCA), Aortic Isthmus (AoI) and Left Ventricular Inflow and Outflow (LVIO)), involving classification and waveform delineation tasks. Derived from data from a low- and middle-income country, our approach's versatility was tested and validated using a dataset from a high-income country, showcasing its potential for standardized and accurate analysis across varied healthcare settings.

RESULTS: The classification of Doppler views was approached through three distinct blocks: (i) a Doppler velocity amplitude-based model with an accuracy of 94%, (ii) two Convolutional Neural Networks (CNN) with accuracies of 89.2% and 67.3%, and (iii) Doppler view- and dataset-dependent confidence models to detect misclassifications with an accuracy higher than 85%. The extraction of Doppler indices utilized Doppler-view dependent CNNs coupled with post-processing techniques. Results yielded a mean absolute percentage error of 6.1 ± 4.9% (n = 682), 1.8 ± 1.5% (n = 1,480), 4.7 ± 4.0% (n = 717), 3.5 ± 3.1% (n = 1,318) for the magnitude location of the systolic peak in LVIO, UA, AoI and MCA views, respectively.

CONCLUSIONS: The developed models proved to be highly accurate in classifying Doppler views and extracting essential measurements from Doppler images. The integration of this AI-enabled workflow holds significant promise in reducing the manual workload and enhancing the efficiency of feto-placental Doppler image analysis, even for non-trained readers.

PMID:39479252 | PMC:PMC11521966 | DOI:10.3389/fdgth.2024.1455767

Front Neurol. 2024 Oct 16;15:1394435. doi: 10.3389/fneur.2024.1394435. eCollection 2024.

ABSTRACT

OBJECTIVES: The diagnosis of intracranial atherosclerotic stenosis (ICAS) is of great significance for the prevention of stroke. Deep learning (DL)-based artificial intelligence techniques may aid in the diagnosis. The study aimed to identify ICAS in the middle cerebral artery (MCA) based on a modified DL model.

METHODS: This retrospective study included two datasets. Dataset1 consisted of 3,068 transcranial Doppler (TCD) images of the MCA from 1,729 patients, which were assessed as normal or stenosis by three physicians with varying levels of experience, in conjunction with other medical imaging data. The data were used to improve and train the VGG16 models. Dataset2 consisted of TCD images of 90 people who underwent physical examination, which were used to verify the robustness of the model and compare the consistency between the model and human physicians.

RESULTS: The accuracy, precision, specificity, sensitivity, and area under curve (AUC) of the best model VGG16 + Squeeze-and-Excitation (SE) + skip connection (SC) on dataset1 reached 85.67 ± 0.43(%),87.23 ± 1.17(%),87.73 ± 1.47(%),83.60 ± 1.60(%), and 0.857 ± 0.004, while those of dataset2 were 93.70 ± 2.80(%),62.65 ± 11.27(%),93.00 ± 3.11(%),100.00 ± 0.00(%), and 0.965 ± 0.016. The kappa coefficient showed that it reached the recognition level of senior doctors.

CONCLUSION: The improved DL model has a good diagnostic effect for MCV stenosis in TCD images and is expected to help in ICAS screening.

PMID:39479004 | PMC:PMC11521853 | DOI:10.3389/fneur.2024.1394435

Artif Intell Rev. 2024;57(12):349. doi: 10.1007/s10462-024-10943-8. Epub 2024 Oct 28.

ABSTRACT

Measuring Efficiency in neural network system development is an open research problem. This paper presents an experimental framework to measure the training efficiency of a neural architecture. To demonstrate our approach, we analyze the training efficiency of Convolutional Neural Networks and Bayesian equivalents on the MNIST and CIFAR-10 tasks. Our results show that training efficiency decays as training progresses and varies across different stopping criteria for a given neural model and learning task. We also find a non-linear relationship between training stopping criteria, training Efficiency, model size, and training Efficiency. Furthermore, we illustrate the potential confounding effects of overtraining on measuring the training efficiency of a neural architecture. Regarding relative training efficiency across different architectures, our results indicate that CNNs are more efficient than BCNNs on both datasets. More generally, as a learning task becomes more complex, the relative difference in training efficiency between different architectures becomes more pronounced.

PMID:39478973 | PMC:PMC11519118 | DOI:10.1007/s10462-024-10943-8

BMC Public Health. 2024 Oct 31;24(1):3014. doi: 10.1186/s12889-024-20532-7.

ABSTRACT

BACKGROUND: Infectious diseases are major medical and social challenges of the 21st century. Accurately predicting incidence is of great significance for public health organizations to prevent the spread of diseases. Internet search engine data, like Baidu search index, may be useful for analyzing epidemics and improving prediction.

METHODS: We collected data on hepatitis E incidence and cases in Shandong province from January 2009 to December 2022 are extracted. Baidu index is available from January 2009 to December 2022. Employing Pearson correlation analysis, we validated the relationship between the Baidu index and hepatitis E incidence. We utilized various LSTM architectures, including LSTM, stacked LSTM, attention-based LSTM, and attention-based stacked LSTM, to forecast hepatitis E incidence both with and without incorporating the Baidu index. Meanwhile, we introduce KAN to LSTM models for improving nonlinear learning capability. The performance of models are evaluated by three standard quality metrics, including root mean square error(RMSE), mean absolute percentage error(MAPE) and mean absolute error(MAE).

RESULTS: Adjusting for the Baidu index altered the correlation between hepatitis E incidence and the Baidu index from -0.1654 to 0.1733. Without Baidu index, we obtained 17.04±0.13%, 17.19±0.57%, in terms of MAPE, by LSTM and attention based stacked LSTM, respectively. With the Baidu index, we obtained 15.36±0.16%, 15.15±0.07%, in term of MAPE, by the same methods. The prediction accuracy increased by 2%. The methods with KAN can improve the performance by 0.3%. More detailed results are shown in results section of this paper.

CONCLUSIONS: Our experiments reveal a weak correlation and similar trends between the Baidu index and hepatitis E incidence. Baidu index proves to be valuable for predicting hepatitis E incidence. Furthermore, stack layers and KAN can also improve the representational ability of LSTM models.

PMID:39478514 | DOI:10.1186/s12889-024-20532-7

BMC Med Imaging. 2024 Oct 30;24(1):294. doi: 10.1186/s12880-024-01478-z.

ABSTRACT

BACKGROUND: We explored whether the feature aggregation and refinement network (FARNet) algorithm accurately identified posteroanterior (PA) cephalometric landmarks.

METHODS: We identified 47 landmarks on 1,431 PA cephalograms of which 1,177 were used for training, 117 for validation, and 137 for testing. A FARNet-based artificial intelligence (AI) algorithm automatically detected the landmarks. Model effectiveness was calculated by deriving the mean radial error (MRE) and the successful detection rates (SDRs) within 2, 2.5, 3, and 4 mm. The Mann-Whitney U test was performed on the Euclidean differences between repeated manual identifications and AI trials. The direction in differences was analyzed, and whether differences moved in the same or opposite directions relative to ground truth on both the x and y-axis.

RESULTS: The AI system (web-based CranioCatch annotation software (Eskişehir, Turkey)) identified 47 anatomical landmarks in PA cephalograms. The right gonion SDRs were the highest, thus 96.4, 97.8, 100, and 100% within 2, 2.5, 3, and 4 mm, respectively. The right gonion MRE was 0.94 ± 0.53 mm. The right condylon SDRs were the lowest, thus 32.8, 45.3, 54.0, and 67.9% within the same thresholds. The right condylon MRE was 3.31 ± 2.25 mm. The AI model's reliability and accuracy were similar to a human expert's. AI was better at four skeleton points than the expert, whereas the expert was better at one skeletal and seven dental points (P < 0.05). Most of the points exhibited significant deviations along the y-axis. Compared to ground truth, most of the points in AI and the second trial showed opposite movement on the x-axis and the same on the y-axis.

CONCLUSIONS: The FARNet algorithm streamlined orthodontic diagnosis.

PMID:39478475 | DOI:10.1186/s12880-024-01478-z

BMC Genomics. 2024 Oct 30;25(1):1019. doi: 10.1186/s12864-024-10954-3.

ABSTRACT

The Golgi apparatus is a crucial component of the inner membrane system in eukaryotic cells, playing a central role in protein biosynthesis. Dysfunction of the Golgi apparatus has been linked to neurodegenerative diseases. Accurate identification of sub-Golgi protein types is therefore essential for developing effective treatments for such diseases. Due to the expensive and time-consuming nature of experimental methods for identifying sub-Golgi protein types, various computational methods have been developed as identification tools. However, the majority of these methods rely solely on neighboring features in the protein sequence and neglect the crucial spatial structure information of the protein.To discover alternative methods for accurately identifying sub-Golgi proteins, we have developed a model called GASIDN. The GASIDN model extracts multi-dimension features by utilizing a 1D convolution module on protein sequences and a graph learning module on contact maps constructed from AlphaFold2.The model utilizes the deep representation learning model SeqVec to initialize protein sequences. GASIDN achieved accuracy values of 98.4% and 96.4% in independent testing and ten-fold cross-validation, respectively, outperforming the majority of previous predictors. To the best of our knowledge, this is the first method that utilizes multi-scale feature fusion to identify and locate sub-Golgi proteins. In order to assess the generalizability and scalability of our model, we conducted experiments to apply it in the identification of proteins from other organelles, including plant vacuoles and peroxisomes. The results obtained from these experiments demonstrated promising outcomes, indicating the effectiveness and versatility of our model. The source code and datasets can be accessed at https://github.com/SJNNNN/GASIDN .

PMID:39478465 | DOI:10.1186/s12864-024-10954-3

BMC Bioinformatics. 2024 Oct 30;25(1):340. doi: 10.1186/s12859-024-05968-3.

ABSTRACT

BACKGROUND: Deep learning-based drug-target affinity (DTA) prediction methods have shown impressive performance, despite a high number of training parameters relative to the available data. Previous studies have highlighted the presence of dataset bias by suggesting that models trained solely on protein or ligand structures may perform similarly to those trained on complex structures. However, these studies did not propose solutions and focused solely on analyzing complex structure-based models. Even when ligands are excluded, protein-only models trained on complex structures still incorporate some ligand information at the binding sites. Therefore, it is unclear whether binding affinity can be accurately predicted using only compound or protein features due to potential dataset bias. In this study, we expanded our analysis to comprehensive databases and investigated dataset bias through compound and protein feature-based methods using multilayer perceptron models. We assessed the impact of this bias on current prediction models and proposed the binding affinity similarity explorer (BASE) web service, which provides bias-reduced datasets.

RESULTS: By analyzing eight binding affinity databases using multilayer perceptron models, we confirmed a bias where the compound-protein binding affinity can be accurately predicted using compound features alone. This bias arises because most compounds show consistent binding affinities due to high sequence or functional similarity among their target proteins. Our Uniform Manifold Approximation and Projection analysis based on compound fingerprints further revealed that low and high variation compounds do not exhibit significant structural differences. This suggests that the primary factor driving the consistent binding affinities is protein similarity rather than compound structure. We addressed this bias by creating datasets with progressively reduced protein similarity between the training and test sets, observing significant changes in model performance. We developed the BASE web service to allow researchers to download and utilize these datasets. Feature importance analysis revealed that previous models heavily relied on protein features. However, using bias-reduced datasets increased the importance of compound and interaction features, enabling a more balanced extraction of key features.

CONCLUSIONS: We propose the BASE web service, providing both the affinity prediction results of existing models and bias-reduced datasets. These resources contribute to the development of generalized and robust predictive models, enhancing the accuracy and reliability of DTA predictions in the drug discovery process. BASE is freely available online at https://synbi2024.kaist.ac.kr/base .

PMID:39478454 | DOI:10.1186/s12859-024-05968-3

Histopathology. 2024 Oct 31. doi: 10.1111/his.15356. Online ahead of print.

ABSTRACT

AIMS: To test the efficacy of artificial intelligence (AI)-assisted Ki-67 digital image analysis in invasive breast carcinoma (IBC) with quantitative assessment of AI model performance.

METHODS AND RESULTS: This study used 494 cases of Ki-67 slide images of IBC core needle biopsies. The methods were divided into two steps: (i) construction of a deep-learning model (DL); and (ii) DL implementation for Ki-67 analysis. First, a DL tissue classifier model (DL-TC) and a DL nuclear detection model (DL-ND) were constructed using HALO AI DenseNet V2 algorithm with 31,924 annotations in 300 Ki-67 digital slide images. Whether the class predicted by DL-TC in the test set was correct compared with the annotation of ground truth at the pixel level was evaluated. Second, DL-TC- and DL-ND-assisted digital image analysis (DL-DIA) was performed in the other 194 luminal-type cases and correlations with manual counting and clinical outcome were investigated to confirm the accuracy and prognostic potential of DL-DIA. The performance of DL-TC was excellent and invasive carcinoma nests were well segmented from other elements (average precision: 0.851; recall: 0.878; F1-score: 0.858). Ki-67 index data and the number of nuclei from DL-DIA were positively correlated with data from manual counting (ρ = 0.961, and 0.928, respectively). High Ki-67 index (cutoff 20%) cases showed significantly worse recurrence-free survival and breast cancer-specific survival (P = 0.024, and 0.032, respectively).

CONCLUSION: The performances of DL-TC and DL-ND were excellent. DL-DIA demonstrated a high degree of concordance with manual counting of Ki-67 and the results of this approach have prognostic potential.

PMID:39478421 | DOI:10.1111/his.15356

Shanghai Kou Qiang Yi Xue. 2024 Aug;33(4):339-344.

ABSTRACT

PURPOSE: The established automatic AI tooth segmentation algorithm was used to achieve rapid and automatic tooth segmentation from CBCT images. The three-dimensional data obtained by oral scanning of real isolated teeth were used as the gold standard to verify the accuracy of the algorithm.

METHODS: Thirty sets of CBCT data and corresponding 59 isolated teeth were collected from Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine. The three-dimensional tooth data in CBCT images were segmented by the algorithm. The digital information obtained by scanning the extracted teeth after processing was used as the gold standard. In order to compare the difference between the segmentation results and the scanning results of the algorithm. The Dice coefficient(Dice), sensitivity (Sen) and average symmetric surface distance (ASSD) were selected to evaluate the segmentation accuracy of the algorithm. The intra-class correlation coefficient(ICC) was used to evaluate the differences in length, area, and volume between the single tooth obtained by the AI system and the digital isolated tooth. Due to the existence of CBCT with different resolution, ANOVA was used to analyze the differences between groups with different resolution, and SNK method was used to compare them between two groups. SPSS 25.0 software package was used to analyze the data.

RESULTS: After comparing the segmentation results with the in vitro dental scanning results, the average Dice value was (94.7±1.88)%, the average Sen was (95.8±2.02)%, and the average ASSD was (0.49±0.12) mm. By comparing the length, area and volume of a single tooth obtained by the digital isolated tooth and the AI system, the ICC values of the intra-group correlation coefficients were 0.734, 0.719 and 0.885, respectively. The single tooth divided by the AI system has a good consistency with the digital model in evaluating the length, area and volume, but the segmentation results were still different from the real situation in terms of specific values. The smaller the voxel of CBCT, the higher the resolution, the better the segmentation results.

CONCLUSIONS: The CBCT tooth segmentation algorithm established in this study can accurately achieve the tooth segmentation of the whole dentition in CBCT at all resolutions. The improvement of CBCT resolution ratio can make the algorithm more accurate. Compared with the current segmentation algorithms, our algorithm has better performance. Compared with the real situation, there are still some differences, and the algorithm needs to be further improved and verified.

PMID:39478388

Adv Sci (Weinh). 2024 Oct 30:e2404448. doi: 10.1002/advs.202404448. Online ahead of print.

ABSTRACT

Miniaturized computational spectrometers have emerged as a promising strategy for miniaturized spectrometers, which breaks the compromise between footprint and performance in traditional miniaturized spectrometers by introducing computational resources. They have attracted widespread attention and a variety of materials, optical structures, and photodetectors are adopted to fabricate computational spectrometers with the cooperation of reconstruction algorithms. Here, a comprehensive review of miniaturized computational spectrometers, focusing on two crucial components: spectral encoding and reconstruction algorithms are provided. Principles, features, and recent progress of spectral encoding strategies are summarized in detail, including space-modulated, time-modulated, and light-source spectral encoding. The reconstruction algorithms are classified into traditional and deep learning algorithms, and they are carefully analyzed based on the mathematical models required for spectral reconstruction. Drawing from the analysis of the two components, cooperations between them are considered, figures of merits for miniaturized computational spectrometers are highlighted, optimization strategies for improving their performance are outlined, and considerations in operating these systems are provided. The application of miniaturized computational spectrometers to achieve hyperspectral imaging is also discussed. Finally, the insights into the potential future applications and developments of computational spectrometers are provided.

PMID:39477813 | DOI:10.1002/advs.202404448