Digestion. 2024 Dec 26:1-13. doi: 10.1159/000543345. Online ahead of print.

ABSTRACT

BACKGROUND: Artificial intelligence (AI) has significantly impacted medical imaging, particularly in gastrointestinal endoscopy. Computer-aided detection and diagnosis systems (CADe and CADx) are thought to enhance the quality of colonoscopy procedures.

SUMMARY: Colonoscopy is essential for colorectal cancer screening, but often misses a significant percentage of adenomas. AI-assisted systems employing deep learning offer improved detection and differentiation of colorectal polyps, potentially increasing adenoma detection rates by 8%-10%. The main benefit of CADe is in detecting small adenomas, whereas it has a limited impact on advanced neoplasm detection. Recent advancements include real-time CADe systems and CADx for histopathological predictions, aiding in the differentiation of neoplastic and non-neoplastic lesions. Biases such as the Hawthorne effect and potential overdiagnosis necessitate large-scale clinical trials to validate the long-term benefits of AI. Additionally, novel concepts such as computer-aided quality improvement systems are emerging to address limitations facing current CADe systems.

KEY MESSAGES: Despite the potential of AI for enhancing colonoscopy outcomes, its effectiveness in reducing colorectal cancer incidence and mortality remains unproven. Further prospective studies are essential to establish the overall utility and clinical benefits of AI in colonoscopy.

PMID:39724867 | DOI:10.1159/000543345

Phys Med. 2024 Dec 25;129:104881. doi: 10.1016/j.ejmp.2024.104881. Online ahead of print.

ABSTRACT

PURPOSE: We investigate the feasibility of using a biophysically guided approach for delineating the Clinical Target Volume (CTV) in Glioblastoma Multiforme (GBM) by evaluating its impact on the treatment outcomes, specifically Overall Survival (OS) time.

METHODS: An established reaction-diffusion model was employed to simulate the spatiotemporal evolution of cancerous regions in T1-MRI images of GBM patients. The effects of the parameters of this model on the simulated tumor borders were quantified and the optimal values were used to estimate the distribution of infiltrative cells (CTVmodel). Radiomics and deep learning models were examined to predict the OS time by analyzing the GTV, clinical CTV, and CTVmodel.

RESULTS: The study involves 126 subjects for model development and 62 independent subjects for testing. Evaluation of the proposed approach demonstrates comparable predictive power for OS time that is achieved with the clinically defined CTV. Specifically, for the binary survival prediction, short vs. long time, the proposed CTVmodelresulted in improvements of prognostic power, in terms of AUROC, both for the validation (0.77 from 0.75) and the testing (0.73 from 0.71) set. Quantitative comparisons for three-class prediction and survival regression models exhibited a similar trend of comparable performance.

CONCLUSION: The findings highlight the potential of biophysical modeling for estimating and incorporating the spread of infiltrating cells into CTV delineation. Further clinical investigations are required to validate the clinical efficacy.

PMID:39724784 | DOI:10.1016/j.ejmp.2024.104881

Invest Radiol. 2024 Oct 9. doi: 10.1097/RLI.0000000000001126. Online ahead of print.

ABSTRACT

PURPOSE: Accurate detection of central venous catheter (CVC) misplacement is crucial for patient safety and effective treatment. Existing artificial intelligence (AI) often grapple with the limitations of label inaccuracies and output interpretations that lack clinician-friendly comprehensibility. This study aims to introduce an approach that employs segmentation of support material and anatomy to enhance the precision and comprehensibility of CVC misplacement detection.

MATERIALS AND METHODS: The study utilized 2 datasets: the publicly accessible RANZCR CLiP dataset and a bespoke in-house dataset of 1006 annotated supine chest x-rays. Three deep learning models were trained: a classification network, a segmentation network, and a combination of both. These models were evaluated using receiver operating characteristic analysis, area under the curve, DICE similarity coefficient, and Hausdorff distance.

RESULTS: The combined model demonstrated superior performance with an area under the curve of 0.99 for correctly positioned CVCs and 0.95 for misplacements. The model maintained high efficacy even with reduced training data from the local dataset. Sensitivity and specificity rates were high, and the model effectively managed the segmentation and classification tasks, even in images with multiple CVCs and other support materials.

CONCLUSIONS: This study illustrates the potential of AI-based models in accurately and reliably determining CVC placement in chest x-rays. The proposed method shows high accuracy and offers improved interpretability, important for clinical decision-making. The findings also highlight the importance of dataset quality and diversity in training AI models for medical image analysis.

PMID:39724590 | DOI:10.1097/RLI.0000000000001126

Int J Surg. 2024 Oct 1. doi: 10.1097/JS9.0000000000002105. Online ahead of print.

ABSTRACT

OBJECTIVE: To develop a model for accurate prediction of axillary lymph node (LN) status after neoadjuvant chemotherapy (NAC) in breast cancer patients with nodal involvement.

METHODS: Between October 2018 and February 2024, 671 breast cancer patients with biopsy-proven LN metastasis who received NAC followed by axillary LN dissection were enrolled in this prospective, multicenter study. Preoperative ultrasound (US) images, including B-mode ultrasound (BUS) and shear wave elastography (SWE), were obtained. The included patients were randomly divided at a ratio of 8:2 into a training set and an independent test set, with five-fold cross-validation applied to training set. We first identified clinicopathological characteristics and conventional US features significantly associated with the axillary LN response and developed corresponding prediction models. We then constructed deep learning radiomics (DLR) models based on BUS and SWE data. Models performances were compared, and a combination model was developed using significant clinicopathological data and interpreted US features with the SWE-based DLR model. Discrimination, calibration and clinical utility of this model were analyzed using receiver operating characteristic curve, calibration curve and decision curve, respectively.

RESULTS: Axillary pathologic complete response (pCR) was achieved in 52.41% of patients. In the test cohort, the clinicopathologic model had an accuracy of 71.30%, while radiologists' diagnoses ranged from 64.26% to 71.11%, indicating limited to moderate predictive ability for the axillary response to NAC. The SWE-based DLR model, with an accuracy of 80.81%, significantly outperformed the BUS-based DLR model, which scored 59.57%. The combination DLR model boasted an accuracy of 88.70% and a false-negative rate of 8.82%. It demonstrated strong discriminatory ability (AUC, 0.95), precise calibration (p value obtained by Hosmer-Lemeshow goodness-of-fit test, 0.68), and practical clinical utility (probability threshold, 2.5-97.5%).

CONCLUSIONS: The combination SWE-based DLR model can predict the axillary status after NAC in patients with node-positive breast cancer, and thus, may inform clinical decision-making to help avoid unnecessary axillary LN dissection.

PMID:39724577 | DOI:10.1097/JS9.0000000000002105

J Chem Inf Model. 2024 Dec 26. doi: 10.1021/acs.jcim.4c01096. Online ahead of print.

ABSTRACT

In recent years, the deep learning (DL) technique has rapidly developed and shown great success in scoring the protein-ligand binding affinities. The protein-ligand conformation optimization based on DL-derived scoring functions holds broad application prospects, for instance, drug design and enzyme engineering. In this study, we evaluated the robustness of a DL-based ligand conformation optimization protocol (DeepRMSD+Vina) for optimizing structures with input perturbations by examining the predicted ligand binding poses and scoring. Our results clearly indicated that compared to traditional optimization algorithms (such as Prime MM-GBSA and Vina optimization), DeepRMSD+Vina exhibits higher performance when treating diverse protein-ligand cases. The DeepRMSD+Vina is robust and can always generate the correct binding structures even if perturbations (up to 3 Å) are introduced to the input structure. The success rate is 62% for perturbation with a RMSD within 2-3 Å. However, the success rate dramatically drops to 11% for large perturbations, with RMSD extending to 3-4 Å. Furthermore, compared to the widely used optimization protocol of AutoDock Vina, the DL-generated conformation shows a balanced performance for all of the systems under examination. Overall, the DL-based DeepRMSD+Vina is unarguably more reliable than the traditional methods, which is attributed to the physically inspired design of the neural networks in DeepRMSD+Vina where the distance-transformed features describing the atomic interactions between the protein and the ligand have been explicitly considered and modeled. The outstanding robustness of the DL-based ligand conformational optimization algorithm further validates its superiority in the field of conformational optimization.

PMID:39724561 | DOI:10.1021/acs.jcim.4c01096

Hum Reprod. 2024 Dec 26:deae267. doi: 10.1093/humrep/deae267. Online ahead of print.

ABSTRACT

STUDY QUESTION: How can we best achieve tissue segmentation and cell counting of multichannel-stained endometriosis sections to understand tissue composition?

SUMMARY ANSWER: A combination of a machine learning-based tissue analysis software for tissue segmentation and a deep learning-based algorithm for segmentation-independent cell identification shows strong performance on the automated histological analysis of endometriosis sections.

WHAT IS KNOWN ALREADY: Endometriosis is characterized by the complex interplay of various cell types and exhibits great variation between patients and endometriosis subtypes.

STUDY DESIGN, SIZE, DURATION: Endometriosis tissue samples of eight patients of different subtypes were obtained during surgery.

PARTICIPANTS/MATERIALS, SETTING, METHODS: Endometriosis tissue was formalin-fixed and paraffin-embedded before sectioning and staining by (multiplex) immunohistochemistry. A 6-plex immunofluorescence panel in combination with a nuclear stain was established following a standardized protocol. This panel enabled the distinction of different tissue structures and dividing cells. Artificial intelligence-based tissue and cell phenotyping were employed to automatically segment the various tissue structures and extract quantitative features.

MAIN RESULTS AND THE ROLE OF CHANCE: An endometriosis-specific multiplex panel comprised of PanCK, CD10, α-SMA, calretinin, CD45, Ki67, and DAPI enabled the distinction of tissue structures in endometriosis. Whereas a machine learning approach enabled a reliable segmentation of tissue substructure, for cell identification, the segmentation-free deep learning-based algorithm was superior.

LIMITATIONS, REASONS FOR CAUTION: The present analysis was conducted on a limited number of samples for method establishment. For further refinement, quantification of collagen-rich cell-free areas should be included which could further enhance the assessment of the extent of fibrotic changes. Moreover, the method should be applied to a larger number of samples to delineate subtype-specific differences.

WIDER IMPLICATIONS OF THE FINDINGS: We demonstrate the great potential of combining multiplex staining and cell phenotyping for endometriosis research. The optimization procedure of the multiplex panel was transferred from a cancer-related project, demonstrating the robustness of the procedure beyond the cancer context. This panel can be employed for larger batch analyses. Furthermore, we demonstrate that the deep learning-based approach is capable of performing cell phenotyping on tissue types that were not part of the training set underlining the potential of the method for heterogenous endometriosis samples.

STUDY FUNDING/COMPETING INTEREST(S): All funding was provided through departmental funds. The authors declare no competing interests.

TRIAL REGISTRATION NUMBER: N/A.

PMID:39724530 | DOI:10.1093/humrep/deae267

Environ Monit Assess. 2024 Dec 26;197(1):103. doi: 10.1007/s10661-024-13595-x.

ABSTRACT

Waste sorting is a key part of sustainable development. To maximize the recovery of resources and reduce labor costs, a waste management and classification system is established. In the system, we use Internet of Things (IoT) and edge computing to implement waste sorting and the systematic long-distance information transmission and monitoring. A dataset of recyclable waste images with realistic backgrounds was collected, where the images contained multiple waste categories in a single image. An improved deep learning model based on YOLOv7-tiny is proposed to adapt to the realistic complex background of waste images. In the model, adding partial convolution (PConv) to Efficient Layer Aggregation Network (ELAN) module reduces parameters and floating point of operations (FLOPs). Coordinate attention (CA) is added to spatial pyramid pooling (Sppcspc) module and ELAN module, respectively. SIoU loss function is used, which improves the recognition accuracy of the model. The improved model shows a higher accuracy on the basis of lighter weight and is more suitable for deployment on edge devices. The proposed model and the original model were trained using our dataset, and their performance was compared. According to the experimental results, mAP@.5, mAP@.5:.95 of the improved YOLOv7-tiny are increased by 1.7% and 1.4%, and the parameter and FLOPs are decreased by 4.8% and 5%, respectively. The improved model has an average inference time of 110 ms and an FPS of 9 on the Jetson Nano. Hence, we believe that the developed system can better meet the needs of current garbage collection system.

PMID:39724392 | DOI:10.1007/s10661-024-13595-x

Genomics Proteomics Bioinformatics. 2024 Dec 26:qzae091. doi: 10.1093/gpbjnl/qzae091. Online ahead of print.

ABSTRACT

Chromatin compartmentalization and epigenomic modification are crucial in cell differentiation and diseases development. However, precise mapping of chromatin compartmental patterns requires Hi-C or Micro-C data at high sequencing depth. Exploring the systematic relationship between epigenomic modifications and compartmental patterns remains challenging. To address these issues, we present COCOA, a deep neural network framework using convolution and attention mechanisms to infer fine-scale chromatin compartment patterns from six histone modification signals. COCOA extracts 1-D track features through bi-directional feature reconstruction after resolution-specific binning epigenomic signals. These track features are then cross-fused with contact features using an attention mechanism and transformed into chromatin compartment patterns through residual feature reduction. COCOA demonstrates accurate inference of chromatin compartmentalization at a fine-scale resolution and exhibits stable performance on test sets. Additionally, we explored the impact of histone modifications on chromatin compartmentalization prediction through in silico epigenomic perturbation experiments. Unlike obscure compartments observed with 1 kb resolution high-depth experimental data, COCOA generates clear and detailed compartmental patterns, highlighting its superior performance. Finally, we demonstrated that COCOA enables cell-type-specific prediction of unrevealed chromatin compartment patterns in various biological processes, making it an effective tool for gaining chromatin compartmentalization insights from epigenomics in diverse biological scenarios. The COCOA python code is publicly available at https://github.com/onlybugs/COCOA.

PMID:39724385 | DOI:10.1093/gpbjnl/qzae091

J Comput Aided Mol Des. 2024 Dec 26;39(1):4. doi: 10.1007/s10822-024-00582-0.

ABSTRACT

Conotoxins, being small disulfide-rich and bioactive peptides, manifest notable pharmacological potential and find extensive applications. However, the exploration of conotoxins' vast molecular space using traditional methods is severely limited, necessitating the urgent need of developing novel approaches. Recently, deep learning (DL)-based methods have advanced to the molecular generation of proteins and peptides. Nevertheless, the limited data and the intricate structure of conotoxins constrain the application of deep learning models in the generation of conotoxins. We propose ConoDL, a framework for the generation and prediction of conotoxins, comprising the end-to-end conotoxin generation model (ConoGen) and the conotoxin prediction model (ConoPred). ConoGen employs transfer learning and a large language model (LLM) to tackle the challenges in conotoxin generation. Meanwhile, ConoPred filters artificial conotoxins generated by ConoGen, narrowing down the scope for subsequent research. A comprehensive evaluation of the peptide properties at both sequence and structure levels indicates that the artificial conotoxins generated by ConoDL exhibit a certain degree of similarity to natural conotoxins. Furthermore, ConoDL has generated artificial conotoxins with novel cysteine scaffolds. Therefore, ConoDL may uncover new cysteine scaffolds and conotoxin molecules, facilitating further exploration of the molecular space of conotoxins and the discovery of pharmacologically active variants.

PMID:39724258 | DOI:10.1007/s10822-024-00582-0

JAMA Oncol. 2024 Dec 26. doi: 10.1001/jamaoncol.2024.5356. Online ahead of print.

ABSTRACT

IMPORTANCE: Only a small fraction of patients with advanced non-small cell lung cancer (NSCLC) respond to immune checkpoint inhibitor (ICI) treatment. For optimal personalized NSCLC care, it is imperative to identify patients who are most likely to benefit from immunotherapy.

OBJECTIVE: To develop a supervised deep learning-based ICI response prediction method; evaluate its performance alongside other known predictive biomarkers; and assess its association with clinical outcomes in patients with advanced NSCLC.

DESIGN, SETTING, AND PARTICIPANTS: This multicenter cohort study developed and independently validated a deep learning-based response stratification model for predicting ICI treatment outcome in patients with advanced NSCLC from whole slide hematoxylin and eosin-stained images. Images for model development and validation were obtained from 1 participating center in the US and 3 in the European Union (EU) from August 2014 to December 2022. Data analyses were performed from September 2022 to May 2024.

EXPOSURE: Monotherapy with ICIs.

MAIN OUTCOMES AND MEASURES: Model performance measured by clinical end points and objective response rate (ORR) differentiation power vs other predictive biomarkers, ie, programmed death-ligand 1 (PD-L1), tumor mutational burden (TMB), and tumor-infiltrating lymphocytes (TILs).

RESULTS: A total of 295 581 image tiles from 958 patients (mean [SD] age, 66.0 [10.6] years; 456 [48%] females and 502 [52%] males) treated with ICI for NSCLC were included in the analysis. The US-based development cohort consisted of 614 patients with median (IQR) follow-up time of 54.5 (38.2-68.1) months, and the EU-based validation cohort, 344 patients with 43.3 (27.4-53.9) months of follow-up. The ORR to ICI was 26% in the developmental cohort and 28% in the validation cohort. The deep learning model's area under the receiver operating characteristic curve (AUC) for ORR was 0.75 (95% CI, 0.64-0.85) in the internal test set and 0.66 (95% CI, 0.60-0.72) in the validation cohort. In a multivariable analysis, the deep learning model's score was an independent predictor of ICI response in the validation cohort for both progression-free (hazard ratio, 0.56; 95% CI, 0.42-0.76; P < .001) and overall survival (hazard ratio, 0.53; 95% CI, 0.39-0.73; P < .001). The tuned deep learning model achieved a higher AUC than TMB, TILs, and PD-L1 in the internal set; in the validation cohort, it was superior to TILs and comparable with PD-L1 (AUC, 0.67; 95% CI, 0.60-0.74), with a 10-percentage point improvement in specificity. In the validation cohort, combining the deep learning model with PD-L1 scores achieved an AUC of 0.70 (95% CI, 0.63-0.76), outperforming either marker alone, with a response rate of 51% compared to 41% for PD-L1 (≥50%) alone.

CONCLUSIONS AND RELEVANCE: The findings of this cohort study demonstrate a strong and independent deep learning-based feature associated with ICI response in patients with NSCLC across various cohorts. Clinical use of this deep learning model could refine treatment precision and better identify patients who are likely to benefit from ICI for treatment of advanced NSCLC.

PMID:39724105 | DOI:10.1001/jamaoncol.2024.5356

PLoS One. 2024 Dec 26;19(12):e0311316. doi: 10.1371/journal.pone.0311316. eCollection 2024.

ABSTRACT

Hereditary angioedema type 1 (HAE1) is a rare, genetically heterogeneous, and autosomal dominant disease. It is a highly variable, insidious, and potentially life-threatening condition, characterized by sudden local, often asymmetric, and episodic subcutaneous and submucosal swelling, caused by pathogenic molecular variants in the SERPING1 gene, which codes for C1-Inhibitor protein. This study performed the phenotypic and molecular characterization of a HAE1 cluster that includes the largest number of affected worldwide. A geographically HAE1 cluster was found in the northeast Colombian department of Boyaca, which accounts for four unrelated families, with 79 suspected to be affected members. Next-Generation Sequencing (NGS) was performed in 2 out of 4 families (Family 1 and Family 4), identifying the variants c.1420C>T and c.1238T>G, respectively. The latter corresponds to a novel mutation. For Families 2 and 3, the c.1417G>A variant was confirmed by Sanger sequencing. This variant had been previously reported to the patient prior to the beginning of this study. Using deep-learning methods, the structure of the C1-Inhibitor protein, p.Gln474* and p.Met413Arg was predicted, and we propose the molecular mechanism related to the etiology of the disease. Using Sanger sequencing, family segregation analysis was performed on 44 individuals belonging to the families analyzed. The identification of this cluster and its molecular analysis will allow the timely identification of new cases and the establishment of adequate treatment strategies. Our results establish the importance of performing population genetic studies in a multi-cluster region for genetic diseases.

PMID:39724085 | DOI:10.1371/journal.pone.0311316

PLoS One. 2024 Dec 26;19(12):e0310417. doi: 10.1371/journal.pone.0310417. eCollection 2024.

ABSTRACT

Deep learning techniques are increasingly being used to classify medical imaging data with high accuracy. Despite this, due to often limited training data, these models can lack sufficient generalizability to predict unseen test data, produced in different domains, with comparable performance. This study focuses on thyroid histopathology image classification and investigates whether a Generative Adversarial Network [GAN], trained with just 156 patient samples, can produce high quality synthetic images to sufficiently augment training data and improve overall model generalizability. Utilizing a StyleGAN2 approach, the generative network produced images with an Fréchet Inception Distance (FID) score of 5.05, matching state-of-the-art GAN results in non-medical domains with comparable dataset sizes. Augmenting the training data with these GAN-generated images increased model generalizability when tested on external data sourced from three separate domains, improving overall precision and AUC by 7.45% and 7.20% respectively compared with a baseline model. Most importantly, this performance improvement was observed on minority class images, tumour subtypes which are known to suffer from high levels of inter-observer variability when classified by trained pathologists.

PMID:39724083 | DOI:10.1371/journal.pone.0310417

Hum Brain Mapp. 2024 Dec 15;45(18):e70104. doi: 10.1002/hbm.70104.

ABSTRACT

White matter hyperintensities (WMH) of presumed vascular origin are a magnetic resonance imaging (MRI)-based biomarker of cerebral small vessel disease (CSVD). WMH are associated with cognitive decline and increased risk of stroke and dementia, and are commonly observed in aging, vascular cognitive impairment, and neurodegenerative diseases. The reliable and rapid measurement of WMH in large-scale multisite clinical studies with heterogeneous patient populations remains challenging, where the diversity of imaging characteristics across studies adds additional complexity to this task. We present segcsvdWMH, a convolutional neural network-based tool developed to provide reliable and accurate WMH quantification across diverse clinical datasets. segcsvdWMH was developed using a large dataset consisting of over 700 fluid-attenuated inversion recovery MRI scans from seven multisite studies, spanning a wide range of clinical populations, WMH burdens, and imaging protocols. Model training incorporated anatomical information through a novel hierarchical segmentation approach, together with extensive data augmentation techniques to improve performance across varied imaging conditions. Benchmarked against three widely available segmentation tools, segcsvdWMH demonstrated superior accuracy, achieving mean Dice score improvements of 7.8% ± 9.7% over HyperMapp3r, 21.8% ± 8.6% over SAMSEG, and 43.5% ± 7.1% over WMH-SynthSeg across four diverse test datasets. segcsvdWMH also maintained consistently high Dice scores across these test datasets (mean DSC = 0.86 ± 0.08), and exhibited strong, stable correlations with periventricular, deep, and total WMH ground truth volumes (mean r = 0.99 ± 0.01). Additionally, segcsvdWMH was robust to low and moderate levels of simulated MRI spike noise artifacts and maintained strong performance across a range of binary segmentation thresholds and WMH burden levels. These findings suggest that segcsvdWMH may provide more accurate and robust WMH segmentation performance for heterogeneous clinical datasets characterized by varying degrees of CSVD severity.

PMID:39723488 | DOI:10.1002/hbm.70104

Front Neurosci. 2024 Dec 11;18:1402657. doi: 10.3389/fnins.2024.1402657. eCollection 2024.

ABSTRACT

INTRODUCTION: Functional magnetic resonance imaging (fMRI) data is highly complex and high-dimensional, capturing signals from regions of interest (ROIs) with intricate correlations. Analyzing such data is particularly challenging, especially in resting-state fMRI, where patterns are less identifiable without task-specific contexts. Nonetheless, interconnections among ROIs provide essential insights into brain activity and exhibit unique characteristics across groups.

METHODS: To address these challenges, we propose an interpretable fusion analytic framework to identify and understand ROI connectivity differences between two groups, revealing their distinctive features. The framework involves three steps: first, constructing ROI-based Functional Connectivity Networks (FCNs) to manage resting-state fMRI data; second, employing a Self-Attention Deep Learning Model (Self-Attn) for binary classification to generate attention distributions encoding group-level differences; and third, utilizing a Latent Space Item-Response Model (LSIRM) to extract group-representative ROI features, visualized on group summary FCNs.

RESULTS: We applied our framework to analyze four types of cognitive impairments, demonstrating their effectiveness in identifying significant ROIs that contribute to the differences between the two disease groups. The results reveal distinct connectivity patterns and unique ROI features, which differentiate cognitive impairments. Specifically, our framework highlighted group-specific differences in functional connectivity, validating its capability to capture meaningful insights from high-dimensional fMRI data.

DISCUSSION: Our novel interpretable fusion analytic framework addresses the challenges of analyzing high-dimensional, resting-state fMRI data. By integrating FCNs, a Self-Attention Deep Learning Model, and LSIRM, the framework provides an innovative approach to discovering ROI connectivity disparities between groups. The attention distribution and group-representative ROI features offer interpretable insights into brain activity patterns and their variations among cognitive impairment groups. This methodology has significant potential to enhance our understanding of cognitive impairments, paving the way for more targeted therapeutic interventions.

PMID:39723421 | PMC:PMC11668745 | DOI:10.3389/fnins.2024.1402657

Environ Health (Wash). 2024 Jul 26;2(12):875-885. doi: 10.1021/envhealth.4c00088. eCollection 2024 Dec 20.

ABSTRACT

The rational design of molecules with the desired functionality presents a significant challenge in chemistry. Moreover, it is worth noting that making chemicals safe and sustainable is crucial to bringing them to the market. To address this, we propose a novel deep learning framework developed explicitly for inverse design of molecules with both functionality and biocompatibility. This innovative approach comprises two predictive models and one generative model, facilitating the targeted screening of novel molecules from created virtual chemical space. Our method's versatility is highlighted in the inverse design process, where it successfully generates molecules with specified motifs or composition, discovers synthetically accessible molecules, and jointly targets functional and safe properties beyond the training regime. The utility of this method is demonstrated in its ability to design ionic liquids (ILs) with enhanced antibacterial properties and reduced cytotoxicity, addressing the issue of balancing functionality and biocompatibility in molecular design.

PMID:39722843 | PMC:PMC11667291 | DOI:10.1021/envhealth.4c00088

Front Radiol. 2024 Dec 11;4:1424065. doi: 10.3389/fradi.2024.1424065. eCollection 2024.

ABSTRACT

Pneumothorax, a life-threatening condition characterized by air accumulation in the pleural cavity, requires early and accurate detection for optimal patient outcomes. Chest X-ray radiographs are a common diagnostic tool due to their speed and affordability. However, detecting pneumothorax can be challenging for radiologists because the sole visual indicator is often a thin displaced pleural line. This research explores deep learning techniques to automate and improve the detection and segmentation of pneumothorax from chest X-ray radiographs. We propose a novel architecture that combines the advantages of fully convolutional neural networks (FCNNs) and Vision Transformers (ViTs) while using only convolutional modules to avoid the quadratic complexity of ViT's self-attention mechanism. This architecture utilizes a patch-based encoder-decoder structure with skip connections to effectively combine high-level and low-level features. Compared to prior research and baseline FCNNs, our model demonstrates significantly higher accuracy in detection and segmentation while maintaining computational efficiency. This is evident on two datasets: (1) the SIIM-ACR Pneumothorax Segmentation dataset and (2) a novel dataset we curated from The Medical City, a private hospital in the Philippines. Ablation studies further reveal that using a mixed Tversky and Focal loss function significantly improves performance compared to using solely the Tversky loss. Our findings suggest our model has the potential to improve diagnostic accuracy and efficiency in pneumothorax detection, potentially aiding radiologists in clinical settings.

PMID:39722784 | PMC:PMC11668597 | DOI:10.3389/fradi.2024.1424065

Cureus. 2024 Dec 25;16(12):e76369. doi: 10.7759/cureus.76369. eCollection 2024 Dec.

ABSTRACT

Background Anemia, a critical public health issue, affects nearly two billion people globally. Frequent monitoring of blood hemoglobin levels is essential for managing its burden. While point-of-care testing (POCT) devices facilitate hemoglobin testing in resource-limited settings, most are invasive and have inherent limitations. The Non-Invasive Anemia Detection App (NiADA) (Monere, UT) offers a non-invasive alternative, utilizing artificial intelligence (AI) to estimate hemoglobin levels from images of the lower eyelid. This low-cost, real-time solution employs a custom computer vision deep-learning algorithm for hemoglobin levels, offering significant potential for early diagnosis and management of anemia. Methods This study evaluated NiADA in two phases. In the first phase, its performance was compared to laboratory measurements and the minimally invasive POCT device, Hemocue Hb 301. In this study, the current version of NiADA version 2 (V2) is also compared against the previous version of NiADA version 1 (V1) to show the improvement in the last six months. In the second phase, NiADA's results were compared against hemoglobin estimations made by a group of medical professionals, as well as against lab analyzers. For both phases, NiADA performance was evaluated using the Bland-Altman plot, regression coefficients, percentage of acceptable limit, Pearson correlation coefficient, and Lin's concordance correlation coefficient. Results The mean difference between NiADA-V2 and laboratory-estimated hemoglobin values was -0.11 g/dL, with limits of agreement (LOA) ranging from +2.86 to -2.64 g/dL, where the upper limit is comparable with HemoCue. The NiADA-V2-acceptable range (percentage of samples falling within ±1 g/dL absolute error) increased to 54% compared to 40% in NiADA-V1. Additionally, NiADA outperformed medical professionals, showing a mean difference of 0.07 g/dL compared to medical professionals' 0.42 g/dL. Conclusion NiADA, as a non-invasive application, exhibits performance comparable to minimally invasive tools and other POCT devices. Its accuracy exceeds that of medical professionals, making it a viable option for anemia screening and monitoring, particularly in community medicine and regions with limited healthcare resources.

PMID:39722662 | PMC:PMC11669323 | DOI:10.7759/cureus.76369

CPT Pharmacometrics Syst Pharmacol. 2024 Dec 25. doi: 10.1002/psp4.13294. Online ahead of print.

ABSTRACT

We employed a mechanistic learning approach, integrating on-treatment tumor kinetics (TK) modeling with various machine learning (ML) models to address the challenge of predicting post-progression survival (PPS)-the duration from the time of documented disease progression to death-and overall survival (OS) in Head and Neck Squamous Cell Carcinoma (HNSCC). We compared the predictive power of model-derived TK parameters versus RECIST and assessed the efficacy of nine TK-OS ML models against conventional survival models. Data from 526 advanced HNSCC patients treated with chemotherapy and cetuximab in the TPExtreme trial were analyzed using a double-exponential model. TK parameters from the first line and maintenance (TKL1) or after four cycles (TK4) were used to predict PPS and post-cycle 4 OS (OS4), combined with 12 baseline parameters. While ML algorithms underperformed compared to the Cox model for PPS, a random survival forest was superior for OS prediction using TK4 and surpassed RECIST-based metrics. This model demonstrated unbiased OS4 prediction, suggesting its potential for improving HNSCC treatment evaluation. Trial Registration: ClinicalTrials.gov identifier: NCT02268695.

PMID:39722558 | DOI:10.1002/psp4.13294

Sheng Wu Gong Cheng Xue Bao. 2024 Dec 25;40(12):4670-4681. doi: 10.13345/j.cjb.240255.

ABSTRACT

The structures and activities of enzymes are influenced by pH of the environment. Understanding and distinguishing the adaptation mechanisms of enzymes to extreme pH values is of great significance for elucidating the molecular mechanisms and promoting the industrial applications of enzymes. In this study, the ESM-2 protein language model was used to encode the secreted microbial proteins with the optimal performance above pH 9 and below pH 5, which yielded 47 725 high-pH protein sequences and 66 079 low-pH protein sequences, respectively. A deep learning model was constructed to identify protein acid-base tolerance based on amino acid sequences. The model showcased significantly higher accuracy than other methods, with the overall accuracy of 94.8%, precision of 91.8%, and a recall rate of 93.4% on the test set. Furthermore, we built a website (https://enzymepred.biodesign.ac.cn), which enabled users to predict the acid-base tolerance by submitting the protein sequences of enzymes. This study has accelerated the application of enzymes in various fields, including biotechnology, pharmaceuticals, and chemicals. It provides a powerful tool for the rapid screening and optimization of industrial enzymes.

PMID:39722525 | DOI:10.13345/j.cjb.240255

Clin Pharmacol Ther. 2024 Dec 25. doi: 10.1002/cpt.3542. Online ahead of print.

ABSTRACT

Since the deep learning revolution of the early 2010s, significant efforts and billions of dollars have been invested in applying artificial intelligence (AI) to drug discovery and development (AIDD). However, despite high expectations, few AI-discovered or AI-designed drugs have entered human clinical trials, and none have achieved clinical approval. In this perspective, we examine the challenges impeding progress and highlight opportunities to harness AI's potential in transforming drug discovery and development.

PMID:39722473 | DOI:10.1002/cpt.3542