Training Set Research Articles

Clinical application and immune infiltration landscape of stemness-related genes in heart failure.

Heart failure (HF) is the leading cause of morbidity and mortality worldwide. Stemness refers to the self-renewal and differentiation ability of cells. However, little is known about the heart's stemness properties. Thus, the current study aims to identify putative stemness-related biomarkers to construct a viable prediction model of HF and characterize the immune infiltration features of HF. HF datasets from the Gene Expression Omnibus (GEO) database were adopted as the training and validation cohorts while stemness-related genes were obtained from GeneCards and previously published papers. Feature selection was performed using two machine learning algorithms. Nomogram models were then constructed to predict HF risk based on the selected key genes. Moreover, the biological functions of the key genes were evaluated using Gene Ontology (GO) and Kyoto Encyclopedia of Genes Genomes (KEGG) pathway analyses, and gene set variation analysis (GSVA) and enrichment analysis (GSEA) were performed between the high- and low-risk groups. The immune infiltration landscape in HF was investigated, and the interaction network of key genes was analysed to predict potential targets and molecular mechanisms. Seven key genes, namely SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3, were included in the proposed nomogram. This nomogram showed good predictive performance for HF diagnosis in the training and validation sets. GO and KEGG analyses revealed that the key genes were primarily associated with ageing, inflammatory processes and DNA oxidation. GSEA and GSVA identified various inflammatory and immune signalling pathways that were enriched between the high- and low-risk groups. The infiltration of 15 immune cell subsets suggests that adaptive immunity has an important role in HF. Our study identified a clinically significant stemness-related signature for predicting HF risk, with the potential to improve early disease diagnosis, optimize risk stratification and provide new strategies for treating patients with HF.

FICE: Text-conditioned fashion-image editing with guided GAN inversion

Fashion-image editing is a challenging computer-vision task where the goal is to incorporate selected apparel into a given input image. Most existing techniques, known as Virtual Try-On methods, deal with this task by first selecting an example image of the desired apparel and then transferring the clothing onto the target person. Conversely, in this paper, we consider editing fashion images with text descriptions. Such an approach has several advantages over example-based virtual try-on techniques: (i) it does not require an image of the target fashion item, and (ii) it allows the expression of a wide variety of visual concepts through the use of natural language. Existing image-editing methods that work with language inputs are heavily constrained by their requirement for training sets with rich attribute annotations or they are only able to handle simple text descriptions. We address these constraints by proposing a novel text-conditioned editing model called FICE (Fashion Image CLIP Editing) that is capable of handling a wide variety of diverse text descriptions to guide the editing procedure. Specifically, with FICE, we extend the common GAN-inversion process by including semantic, pose-related, and image-level constraints when generating images. We leverage the capabilities of the CLIP model to enforce the text-provided semantics, due to its impressive image–text association capabilities. We furthermore propose a latent-code regularization technique that provides the means to better control the fidelity of the synthesized images. We validate the FICE through rigorous experiments on a combination of VITON images and Fashion-Gen text descriptions and in comparison with several state-of-the-art, text-conditioned, image-editing approaches. Experimental results demonstrate that the FICE generates very realistic fashion images and leads to better editing than existing, competing approaches. The source code is publicly available from: https://github.com/MartinPernus/FICE.

Improved Monitoring and Classification of Engine Oil Condition through Two Machine Learning Techniques

<div>This study explores the effectiveness of two machine learning models, namely multilayer perceptron neural networks (MLP-NN) and adaptive neuro-fuzzy inference systems (ANFIS), in advancing maintenance management based on engine oil analysis. Data obtained from a Mercedes Benz 2628 diesel engine were utilized to both train and assess the MLP-NN and ANFIS models. Six indices—Fe, Pb, Al, Cr, Si, and PQ—were employed as inputs to predict and classify engine conditions. Remarkably, both models exhibited high accuracy, achieving an average precision of 94%. While the radial basis function (RBF) model, as presented in a referenced article, surpassed ANFIS, this comparison underscored the transformative potential of artificial intelligence (AI) tools in the realm of maintenance management. Serving as a proof-of-concept for AI applications in maintenance management, this study encourages industry stakeholders to explore analogous methodologies.</div> <section> <h2>Highlights</h2> <div> <ul> <li> <div>Two machine learning models, multilayer perceptron neural networks (MLP-NN) and adaptive neuro-fuzzy inference systems (ANFIS), were employed to predict and classify the performance condition of diesel engines.</div> </li> <li> <div>Among various training algorithms, Levenberg–Marquardt and the Bayesian regularization demonstrated superior classification accuracy, achieving a 95%–96% range.</div> </li> <li> <div>To assess the generalizability of MLP-NN and ANFIS, the training set size was varied from 90% to 10%.</div> </li> <li> <div>The ANFIS model exhibited greater stability than MLP-NN, with a 50% higher performance.</div> </li> </ul> </div> </section> <section> <h2>Graphical Abstract</h2> <div> <img/> </div> </section>

Development and external validation of a novel modality for rapid recognition of aortic dissection based on peripheral pulse oximetry waveforms.

Aortic dissection (AD) is a life-threatening cardiovascular emergency that is often misdiagnosed as other chest pain conditions. Physiologically, AD may cause abnormalities in peripheral blood flow, which can be detected using pulse oximetry waveforms. This study aimed to assess the feasibility of identifying AD based on pulse oximetry waveforms and to highlight the key waveform features that play a crucial role in this diagnostic method. This prospective study employed high-risk chest pain cohorts from two emergency departments. The initial cohort was enriched with AD patients (n=258, 47% AD) for model development, while the second cohort consisted of chest pain patients awaiting angiography (n=71, 25% AD) and was used for external validation. Pulse oximetry waveforms from the four extremities were collected for each patient. After data preprocessing, a recognition model based on the random forest algorithm was trained using patients' gender, age, and waveform difference features extracted from the pulse oximetry waveforms. The performance of the model was evaluated using receiver operating characteristic (ROC) curve analysis and decision curve analysis (DCA). The importance of features was also assessed using Shapley Value and Gini importance. The model demonstrated strong performance in identifying AD in both the training and external validation sets. In the training set, the model achieved an area under the ROC curve of 0.979 (95% CI: 0.961-0.990), sensitivity of 0.918 (95% CI: 0.873-0.955), specificity of 0.949 (95% CI: 0.912-0.985), and accuracy of 0.933 (95% CI: 0.904-0.959). In the external validation set, the model attained an area under the ROC curve of 0.855 (95% CI: 0.720-0.965), sensitivity of 0.889 (95% CI: 0.722-1.000), specificity of 0.698 (95% CI: 0.566-0.812), and accuracy of 0.794 (95% CI: 0.672-0.878). Decision curve analysis (DCA) further showed that the model provided a substantial net benefit for identifying AD. The median mean and median variance of the four limbs' signals were the most influential features in the recognition model. This study demonstrated the feasibility and strong performance of identifying AD based on peripheral pulse oximetry waveforms in high-risk chest pain populations in the emergency setting. The findings also provided valuable insights for future human fluid dynamics simulations to elucidate the impact of AD on blood flow in greater detail.

Magnetic resonance imaging-based radiomics for predicting infiltration levels of CD68+ tumor-associated macrophages in glioblastomas.

Tumor-associated macrophages (TAMs) are important biomarkers of tumor invasion and prognosis in patients with glioblastoma. We combined the imaging and radiomics features of preoperative MRI to predict CD68+ macrophage infiltration. Clinical, MRI image, and pathology data of 188 patients with glioblastoma were analyzed. Overall, 143 patients were included in the training (n = 101) and validation (n = 42) sets, whereas 45patients were included in an independent test set. The optimal cut-off value (14.8%) was based on the minimum p-value formed by the Kaplan-Meier survival analysis and log-rank tests which divided patients into groups with high CD68+ TAMs(≥ 14.8%) and low CD68+ TAMs (< 14.8%). Regions of interest and radiomics features extraction were based on contrast-enhanced T1-weighted images (CE-T1WI) and T2WI. Multi-parameter stepwise regression was used to create the clinical, radiomics, and combined models, each evaluated using the receiver operating characteristic curve. Decision curve analysis was used to assess the clinical applicability of the nomogram. Aclinical model based on the minimum apparent diffusion coefficient (ADCmin) revealed an area under the curve (AUC) of 0.768, 0.764, and 0.624 for the training set, validation set, and test set, respectively. The 2D radiomics model, based on two features, revealed an AUC of 0.783, 0.724, and 0.789 for the training, validation, and test sets, respectively. The 3D radiomics model, based on three features, revealed AUCs of 0.823, 0.811, and 0.787 for the training, validation, and test sets, respectively. The combined model, with ADCmin and radiomics features, showed the best performance, with AUCs of 0.865, 0.822, and 0.776 for the training, validation, and test sets, respectively. The calibration curve of the combined model nomogram showed good agreement between the estimated and actual probabilities. The combined model constructed using ADCmin, aquantitative imaging parameter, combined with five key radiomics features can be used to evaluate the extent of CD68+ macrophages before surgery.

Predicting prognosis for epithelial ovarian cancer patients receiving bevacizumab treatment with CT-based deep learning

Epithelial ovarian cancer (EOC) presents considerable difficulties in prognostication and treatment strategy development. Bevacizumab, an anti-angiogenic medication, has demonstrated potential in enhancing progression-free survival (PFS) in EOC patients. Nevertheless, the identification of individuals at elevated risk of disease progression following treatment remains a challenging task. This study was to develop and validate a deep learning (DL) model using retrospectively collected computed tomography (CT) plain scans of inoperable and recurrent EOC patients receiving bevacizumab treatment diagnosed between January 2013 and January 2024. A total of 525 patients from three different institutions were retrospectively included in the study and divided into training set (N = 400), internal test set (N = 97) and external test set (N = 28). The model’s performance was evaluated using Harrell’s C-index. Patients were categorized into high-risk and low-risk group based on a predetermined cutoff in the training set. Additionally, a multimodal model was evaluated, incorporating the risk score generated by the DL model and the pretreatment level of carbohydrate antigen 125 as input variables. The Net Reclassification Improvement (NRI) metric quantified the reclassification performance of our optimal model in comparison to the International Federation of Gynecology and Obstetrics (FIGO) staging model. The results indicated that DL model achieved a PFS predictive C-index of 0.73 in the internal test set and a C-index of 0.61 in the external test set, along with hazard ratios of 34.24 in the training set (95% CI: 21.7, 54.1; P < 0.001) and 8.16 in the internal test set (95% CI: 2.5, 26.8; P < 0.001). The multimodal model demonstrated a C-index of 0.76 in the internal test set and a C-index of 0.64 in the external test set. Comparative analysis against FIGO staging revealed an NRI of 0.06 (P < 0.001) for the multimodal model. The model presents opportunities for prognostic assessment, treatment strategizing, and ongoing patient monitoring.

Spectral-based estimation of chlorophyll content and determination of background interference mechanisms in low-coverage rice

The chlorophyll content is a vital indicator of rice growth and nutritional status. However, estimating the rice chlorophyll content using spectral-based techniques at the early tillering stage is challenging because of background interference. Using the energy conservation principle, this study explained the spectral variation and background interference mechanisms of clear, muddy, and green algae-covered backgrounds. We developed mathematical interference models for the three types of backgrounds and determined their interference degree and influence mode. We developed rice chlorophyll content estimation models for unclassified and classified (clear, muddy, and green algae-covered) backgrounds using 12 preprocessing, four wavelength selection, and three modeling methods, and we explored the importance of background classification. Moreover, we found that the optimal chlorophyll content estimation model for the clear background was SS+UVE+CNN, with R2 and RMSE values of 0.786 and 13.191 in the training set and 0.741 and 15.327 in the test set, respectively; that for the muddy background was MSC+GA+CNN, with R2 and RMSE values of 0.914 and 10.425 in the training set and 0.660 and 16.844 in the test set, respectively; and that for the green algae-covered background was DC+GA+CNN, with R2 and RMSE values of 0.904 and 9.111 in the training set and 0.688 and 17.694 in the test set, respectively. Our study could provide valuable insights into reducing and correcting background interference during proximal remote sensing data collection.

Construction and validation of a risk prediction model for soldiers with frostbite in northeast China: a cross-sectional study

BackgroundOne of the challenges of physical training in extreme condition is frostbite, especially in Northeast China. In this study, we aimed to construct a risk prediction model for frostbite among soldiers in Northeast China, and verify its effect.Methods698 participants were selected via convenience sampling from Northeast China from December 2021 to January 2022 (winter). They were randomly divided into a training set (N = 479) and a testing set (N = 202) in a ratio of 7:3. All participants completed a researcher-made questionnaire on frostbite. The prediction model was constructed through the use of Logistic regression analysis, which was used to predict the independent risk factors for frostbite formation and screen significant indicators. The model’s performance was assessed using the receiver operating characteristic (ROC) curve and decision curve analysis (DCA) to evaluate the prediction efficiency and goodness of fit.ResultsThe incidence of frostbite in the training set was 19.83% (95 people), all of which were first-degree frostbite. Among them, frostbite in multiple parts was the most common (58.95%), followed by singular body parts like hands (24.21%), ears (11.58%) and feet (5.26%). Single factor logistic regression analyses showed that ambient temperature, ambient wind speed, outdoor stationary time, stationary status, and history of frostbite are independent risk factors that affect the occurrence of frostbite. Furthermore, we constructed the frostbite risk prediction model for soldiers in the northeastern region of China. The area under the receiver operating characteristic curve (AUC) for the risk of frostbite in the training set and testing set was 0.816 (95% CI, 0.770 ~ 0.862) and 0.787 (95% CI, 0.713 ~ 0.860), respectively. The Hosmer-Lemeshow test of the model showed χ2 = 11.328 and P = 0.184 (> 0.05). The DCA curve indicated that most of the clinical net benefits of the model are greater than 0, demonstrating good clinical usefulness.ConclusionThe constructed frostbite prediction model can effectively identify soldiers with a higher risk of frostbite. It provided theoretical support for commanders to take preventive measures to reduce the incidence of frostbite among soldiers and was of great clinical guiding significance.

67 Precision identification and weight assessment of cattle using supervised machine learning on body surface keypoints

Abstract Computer vision systems (CVS) offer identification solutions for animals with distinct coat patterns, but are less effective for solid-colored herds. In addition, they can be used to measure critical phenotypes, such as body weight (BW). Providing both BW and identification for solid-colored animals can help farmers make decisions. This study aimed to 1) develop an automatic CVS capable of identifying using the Euclidean distance between keypoints located at specific anatomical landmarks (e.g., bony prominences), and 2) predict the BW using features extracted from these keypoints. The keypoint model was trained to identify seven keypoints per cow, located at the left and right hips, left and right pin bones, tail head, sacral, and cervical vertebrae. These locations remain constant as body condition score of a cow changes. The Euclidean distance between these keypoints was used to generate biometric features for each cow. The keypoint detection model was trained using 3,928 images over 900 epochs with a batch size of 14 and validated using 391 images. This trained model was used to predict keypoints on a different set of 41 cows over 5 d, resulting in a total of 6,944 images which correspond to the dataset for cow identification. The Random Forest (RF) algorithm for cow identification was trained with the first 4 d (5,376 images), and the last day served as the test set (1,568 images). For the cow identification, a RF algorithm was employed on the Euclidean distance of the keypoints. The hyperparameters were determined through a grid search on the training set, utilizing 5-fold cross-validation. The hyperparameters ‘mtry’ (which ranged from 1 to 10) and ‘ntree’ (which ranged from 250 to 500, in increments of 50) were optimized. For the second objective, due to the completely new environment, the keypoint model was fine-tuned using transfer learning for 240 epochs with a batch size of 14, utilizing 310 images. The model predicted keypoints in 1,593 images from different beef-on-dairy animals. This dataset was used to train the BW prediction model. The Euclidean distances from the predicted images and sex were used as features. A RF model was trained using a leave-one-animal-out cross-validation approach. The hyperparameters were determined through a grid search on the training set, using 10-fold cross-validation. The ‘mtry’ parameter ranged from 2 to 16, in increments of 2. For animal identification, the model achieved an accuracy, precision, recall, and F1-score of 92.7%, 89.0%, 90.2%, and 92.7%, respectively. To predict BW, the RF model achieved an R² of 0.86 and a root mean squared prediction error of 36.9 kg, representing 6.7% of the observed values average. These results suggest that keypoints located on the dorsal body surface can identify and weigh individual animals, even those lacking distinct coat color patterns.

Prediction of axillary lymph node metastasis using a magnetic resonance imaging radiomics model of invasive breast cancer primary tumor

BackgroundThis study investigated the clinical value of breast magnetic resonance imaging (MRI) radiomics for predicting axillary lymph node metastasis (ALNM) and to compare the discriminative abilities of different combinations of MRI sequences.MethodsThis study included 141 patients diagnosed with invasive breast cancer from two centers (center 1: n = 101, center 2: n = 40). Patients from center 1 were randomly divided into training set and test set 1. Patients from center 2 were assigned to the test set 2. All participants underwent preoperative MRI, and four distinct MRI sequences were obtained. The volume of interest (VOI) of the breast tumor was delineated on the dynamic contrast-enhanced (DCE) postcontrast phase 2 sequence, and the VOIs of other sequences were adjusted when required. Subsequently, radiomics features were extracted from the VOIs using an open-source package. Both single- and multisequence radiomics models were constructed using the logistic regression method in the training set. The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and precision of the radiomics model for the test set 1 and test set 2 were calculated. Finally, the diagnostic performance of each model was compared with the diagnostic level of junior and senior radiologists.ResultsThe single-sequence ALNM classifier derived from DCE postcontrast phase 1 had the best performance for both test set 1 (AUC = 0.891) and test set 2 (AUC = 0.619). The best-performing multisequence ALNM classifiers for both test set 1 (AUC = 0.910) and test set 2 (AUC = 0.717) were generated from DCE postcontrast phase 1, T2-weighted imaging, and diffusion-weighted imaging single-sequence ALNM classifiers. Both had a higher diagnostic level than the junior and senior radiologists.ConclusionsThe combination of DCE postcontrast phase 1, T2-weighted imaging, and diffusion-weighted imaging radiomics features had the best performance in predicting ALNM from breast cancer. Our study presents a well-performing and noninvasive tool for ALNM prediction in patients with breast cancer.

80 Empowering informed choices: How computer vision can assist consumers in making decisions about meat quality

Abstract Consumer perception of meat quality is highly influenced by meat tenderness and juiciness. Artificial intelligence technology, such as computer vision systems (CVS), can be a powerful tool to predict such meat attributes in retail markets and help consumers make informed decisions. This study aimed to develop a CVS using smartphone images to classify beef steaks and pork chops tenderness (1), predicting shear-force (SF) and intramuscular fat (IMF) content (2), and performing a comparative evaluation between consumer assessments and the CVS (3). The dataset for beef steak consisted of 915 images (one image per steak), with 691 in the training set, 82 in the validation set, and 142 in the testing set. For pork, there were 514 images (one image per chop), distributed as 377 in the training set, 80 in the validation set, and 57 in the testing set. We trained deep neural networks (Xception), pretrained on the ImageNet dataset, for image classification and regression. The tenderness of the steaks was categorized based on Warner-Bratzler shear-force (N) as tender, intermediate, and tough. To achieve the third object, 1,000 pairwise comparisons were drawn from the 142 testing set images, where both consumers and the CVS performed predictions on tender and tough meat. The option with the least SF in each pair was identified as the most tender meat, with ground truth established using SF measurements. The same pairs were presented to meat consumers through a survey. For classifying beef as tender and tough, the algorithms demonstrated F1-scores of 68.1% and 70.8%, respectively. However, intermediate meat yielded worse results, with an F1-score of 36.4%. The dataset was subsequently re-categorized into two classes: tender and tough, with all intermediate steaks classified as tough, and resulted in an F1-score of 76.6% for tough meat. For classifying pork chops as tender and tough, the algorithms demonstrated F1-scores of 81.4% and 85.7%, respectively. The intermediate class in pork chops also exhibited worse results, with an F1-score of 24.0%. After re-categorizing the dataset into two classes, the F1-score for classifying pork chops as tough was 83.9%. The regression model predicted SF in beef steak and pork chops with R2 of 0.64 and 0.76, and RMSEP of 16.9 N and 9.15 N, respectively. The regression model, after analyzing with 1,000 pairwise comparisons, accurately predicted the tenderest steak with an accuracy of 76.5%. In contrast, human recognition of meat tenderness achieved an average accuracy of 46.7% for beef steak. To predict IMF in beef steaks and pork chops the model achieved R2 values of 0.61 and 0.54, and RMSEP of 2.6% and 1.22%, respectively. These findings suggest that CVS can provide a more objective method for evaluating meat tenderness and IMF before purchase, potentially enhancing consumer satisfaction.

HN-SIB-BPI: A single click, sub-site specific, dosimetric scorecard tuned RapidPlan model created from a foundation model for treating head and neck with bilateral neck

To develop a Knowledge Based Planning (KBP) model for creating quantifiably high quality VMAT treatment plans in a single click for head and neck cases treated Simultaneous Integrated Boost (SIB) with bilateral parotid involvement (BPI) where both parotids are near, abutting or partially overlapping target volume. Eclipse RapidPlan and the publicly available PlanScorecard tool were used to assess existing Head and Neck RapidPlan models on two representative cases. The best performer was used as a foundation model to assist in creating new initial training set doses from previously treated cases. Those initial 27 cases were first replanned using only the selected foundation model, then further improved based on manual replanning, informed by dosimetric scorecard assessment. A new, initial model was trained from those 27 foundation model created cases that had been manually improved. Then, that initial model was used to replan those cases again, resulting in higher scores. Additional cases were also replanned using the initial model along with some manual changes to the optimization objectives to increase the score. This resulted in a total of 66 cases from which the final, released, HN-SIB-BPI was trained. A 27 case subset of the full training set was replanned and rescored at each phase of the process with a 260 total point 3-target scorecard. The average score increased: 210.5 foundation model; 226.96 manually improved plans; 230.1 initial model; 231.7 HN-SIB-BPI. On the same 27 case subset, mean ipsilateral and contralateral parotid dose decreased by 1.05Gy and 1.58Gy respectively from the foundation model to HN-SIB-BP. Eight external cases were created from HN-SIB-BPI with dosimetric scorecard validation on Halcyon (3-PTV:221.38/260; 2-PTV:196.1/228.5) and TrueBeam (3-PTV:222.01/260; 2-PTV:202.24/228.5). A specific clinical intent (ie: max parotid sparing) can be articulated in a comprehensive and precise manner by creating a dosimetric scorecard with individual metrics points assigned to each OAR and target metric reflecting their relative importance. This process improved the KBP model (HN-SIB-BPI) in several quantifiable ways including further sparing of parotid dose. All results and tools in this work are shared publicly.

Spatial resolution improvement method of RDTS assisted by small-scale thermal region length recognition model

The Total Variation Deconvolution (TVD) algorithm plays an important role in signal reconstruction, however, when it is used to improve the spatial resolution of Raman Distributed Temperature Sensor (RDTS), there are certain challenges in parameter settings. This paper proposes to use Fully-Connected Neural Network to identify the length of small-scale thermal regions(SSTR), and based on the recognition results to set the TVD parameters automatically. We constructed training sets based on the periodic changes of SSTR signals in RDTS (which we call Thermal Region Response Modes, TRRM), to verify performance, we conducted comparative experiments between models obtained from a training set containing 100 types of TRRMs and 25 types of TRRMs, the Macro-F1 value of the former one is 0.2749 higher, reaching 0.9087, performed well in SSTR length recognition tasks. the traditional TVD assisted by this model can increase the spatial resolution of RDTS from 1.6 m to 0.4 m without manual intervention, which complements the lack of automation in applications of TVD and has practical value.

Can GPT-3.5 generate and code discharge summaries?

The aim of this study was to investigate GPT-3.5 in generating and coding medical documents with International Classification of Diseases (ICD)-10 codes for data augmentation on low-resource labels. Employing GPT-3.5 we generated and coded 9606 discharge summaries based on lists of ICD-10 code descriptions of patients with infrequent (or generation) codes within the MIMIC-IV dataset. Combined with the baseline training set, this formed an augmented training set. Neural coding models were trained on baseline and augmented data and evaluated on an MIMIC-IV test set. We report micro- and macro-F1 scores on the full codeset, generation codes, and their families. Weak Hierarchical Confusion Matrices determined within-family and outside-of-family coding errors in the latter codesets. The coding performance of GPT-3.5 was evaluated on prompt-guided self-generated data and real MIMIC-IV data. Clinicians evaluated the clinical acceptability of the generated documents. Data augmentation results in slightly lower overall model performance but improves performance for the generation candidate codes and their families, including 1 absent from the baseline training data. Augmented models display lower out-of-family error rates. GPT-3.5 identifies ICD-10 codes by their prompted descriptions but underperforms on real data. Evaluators highlight the correctness of generated concepts while suffering in variety, supporting information, and narrative. While GPT-3.5 alone given our prompt setting is unsuitable for ICD-10 coding, it supports data augmentation for training neural models. Augmentation positively affects generation code families but mainly benefits codes with existing examples. Augmentation reduces out-of-family errors. Documents generated by GPT-3.5 state prompted concepts correctly but lack variety, and authenticity in narratives.

Main challenges on the curation of large scale datasets for pancreas segmentation using deep learning in multi-phase CT scans: Focus on cardinality, manual refinement, and annotation quality

Accurate segmentation of the pancreas in computed tomography (CT) holds paramount importance in diagnostics, surgical planning, and interventions. Recent studies have proposed supervised deep-learning models for segmentation, but their efficacy relies on the quality and quantity of the training data. Most of such works employed small-scale public datasets, without proving the efficacy of generalization to external datasets. This study explored the optimization of pancreas segmentation accuracy by pinpointing the ideal dataset size, understanding resource implications, examining manual refinement impact, and assessing the influence of anatomical subregions. We present the AIMS-1300 dataset encompassing 1,300 CT scans. Its manual annotation by medical experts required 938 h. A 2.5D UNet was implemented to assess the impact of training sample size on segmentation accuracy by partitioning the original AIMS-1300 dataset into 11 smaller subsets of progressively increasing numerosity. The findings revealed that training sets exceeding 440 CTs did not lead to better segmentation performance. In contrast, nnU-Net and UNet with Attention Gate reached a plateau for 585 CTs. Tests on generalization on the publicly available AMOS-CT dataset confirmed this outcome. As the size of the partition of the AIMS-1300 training set increases, the number of error slices decreases, reaching a minimum with 730 and 440 CTs, for AIMS-1300 and AMOS-CT datasets, respectively. Segmentation metrics on the AIMS-1300 and AMOS-CT datasets improved more on the head than the body and tail of the pancreas as the dataset size increased. By carefully considering the task and the characteristics of the available data, researchers can develop deep learning models without sacrificing performance even with limited data. This could accelerate developing and deploying artificial intelligence tools for pancreas surgery and other surgical data science applications.

Predicting the risk of nursing home placement of elderly persons using a population-based stratification score

ObjectiveTo develop and validate a novel score predictive of nursing home placement in elderly. Study designPopulation-based case-control study based on healthcare utilization databases of Lombardy, a region of Northern Italy. MethodsThe 2.4 million citizens aged ≥65 years who on January 1, 2018 lived outside nursing home formed the target population. Cases were citizens who experienced nursing home admission (the outcome of interest) until December 31, 2019. Cases were matched 1:1 by gender, age, and municipality of residence to one control. Conditional logistic regression was fitted to select candidate predictors (the exposure to 69 clinical conditions and 11 social and healthcare services) independently associated with the outcome. The model was built from the 26,156 cases, and as many controls (training set), and applied to a validation set (15,807 case-control couples). Predictive performance was assessed by discrimination and calibration. ResultsTwenty-one factors were identified as predictive of nursing home admission and were included in the “Elderly Nursing Home Placement” (ENHP) score. Mental health disorders and chronic neurological illnesses contributed most to prediction of nursing home admission. ENHP performance showed an area under the receiver operating characteristic curve of 0.77 and a remarkable calibration of observed and predicted outcome risk. ConclusionsA simple score derived from data used for public health management may reliably predict the risk of nursing home placement in elderly. Its use by healthcare decision makers allows to accurately identify high-risk individuals who need home services, thereby avoiding admission to nursing homes.

Spot-checking machine learning algorithms for tool wear monitoring in automatic drilling operations in CFRP/Ti6Al4V/Al stacks in the aircraft industry

In aircraft manufacturing, where diverse materials, including Carbon Fiber-Reinforced Plastics (CFRP), aluminum, and titanium alloys, are employed, the assembly process heavily relies on creating thousands of holes. These holes accommodate bolts and rivets, facilitating the secure interlocking of structural components within the aircraft fuselage. The proliferation of sensor systems in this domain has led to a substantial increase in data generation during the hole-making process, offering a compelling opportunity to optimize the production system. In this context, this article is dedicated to harnessing the data collected from the production system of a commercial aircraft to refine the assembly process, with a specific focus on reducing consumable costs. The primary approach involves developing a real-time Tool Wear Monitoring System by comparing the performance of Linear Regression, Lasso Regression, Ridge Regression, k-Nearest Neighbors, Support Vector Regression, Decision Tree, Random Forest, and Extreme Gradient Boosting Machine Learning models. Using a scale of the general drill condition as an outcome, the Gradient Boosting Regressor has shown outstanding results. Notably, the residuals consistently exhibited zero-centered errors in training and test sets. However, it suggests that further enhancements are needed to surpass human-level performance in predicting tool conditions because of the quality and quantity of available data.

Exploring clinical factors to predict the survival for resectable non-small cell lung cancer patients with neoadjuvant immunotherapy.

To explore clinical factors and build a predictive model for the disease-free and overall survival in non-small cell lung cancer patients receiving neoadjuvant chemotherapy combined with immune checkpoint inhibitors. Inclusion criteria for patients in this multicentre study were: (1) patients were diagnosed with stage I-III non-small cell lung cancer diagnosed by bronchoscopy biopsy or puncture; (2) computed tomography/positron emission tomography-computed tomography was applied before treatment and surgery; (3) neoadjuvant chemotherapy combined with immune checkpoint inhibitors were applied for 2-6 cycles preoperatively; (4) peripheral blood indicators and tumour markers were assessed before treatment and surgery; (5) patients underwent radical lung cancer surgery after neoadjuvant therapy. Cases were divided into high- and low-risk groups according to 78 clinical indicators based on 10-fold LASSO selection. We employed Cox proportional hazards models in predicting disease-free and overall survival. Then, we used time-dependent area under the curve and decision curve analyses to examine the accuracy of the results. Data were collected continuously, and 212 and 85 cases were randomly assigned to training and testing sets, respectively. The area under curve for the prediction of disease-free survival (training-1-year, 0.83; 2-year, 0.81; 3-year, 0.83 vs testing-1-year, 0.65; 2-year, 0.66; 3-year, 0.70), overall survival (training-1-year, 0.86; 2-year, 0.85; 3-year, 0.86 vs testing-1-year, 0.66; 2-year, 0.57; 3-year, 0.70) were determined. The coefficient factors including pathological response, preoperative tumour maximum diameter, preoperative lymph shorter-diameter, preoperative tumour&lymph maximum standardized uptake value, change in tumour standardized uptake value preoperative, and blood related risk factors were favorably associated with prognosis (P < 0.001). Our prediction model integrating data from preoperative positron emission tomography-CT, preoperative blood parameters, and pathological response was able to make high accuracy predictions for disease-free and overall survival in non-small cell lung cancer patients receiving neoadjuvant immunity with chemical therapy.

Tumor-Resident Microbiota-Based Risk Model Predicts Neoadjuvant Therapy Response of Locally Advanced Esophageal Squamous Cell Carcinoma Patients.

Few predictive biomarkers exist for identifying patients who may benefit from neoadjuvant therapy (NAT). The intratumoral microbial composition is comprehensively profiled to predict the efficacy and prognosis of patients with esophageal squamous cell carcinoma (ESCC) who underwent NAT and curative esophagectomy. Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis is conducted to screen for the most closely related microbiota and develop a microbiota-based risk prediction (MRP) model on the genera of TM7x, Sphingobacterium, and Prevotella. The predictive accuracy and prognostic value of the MRP model across multiple centers are validated. The MRP model demonstrates good predictive accuracy for therapeutic responses in the training, validation, and independent validation sets. The MRP model also predicts disease-free survival (p= 0.00074 in the internal validation set and p= 0.0017 in the independent validation set) and overall survival (p= 0.00023 in the internal validation set and p= 0.11 in the independent validation set) of patients. The MRP-plus model basing on MRP, tumor stage, and tumor size can also predict the patients who can benefit from NAT. In conclusion, the developed MRP and MRP-plus models may function as promising biomarkers and prognostic indicators accessible at the time of diagnosis.

Machine Learning Models to Predict Early Breakthrough of Recalcitrant Organic Micropollutants in Granular Activated Carbon Adsorbers.

Granular activated carbon (GAC) adsorption is frequently used to remove recalcitrant organic micropollutants (MPs) from water. The overarching aim of this research was to develop machine learning (ML) models to predict GAC performance from adsorbent, adsorbate, and background water matrix properties. For model calibration, MP breakthrough curves were compiled and analyzed to determine the bed volumes of water that can be treated until MP breakthrough reaches ten percent of the influent MP concentration (BV10). Over 400 data points were split into training, validation, and testing sets. Seventeen variables describing MP, background water matrix, and GAC properties were explored in ML models to predict log10-transformed BV10 values. Using the ML models on the testing set, predicted BV10 values exhibited mean absolute errors of ∼0.12 log units and were highly correlated with experimentally determined values (R2 ≥ 0.88). The top three drivers influencing BV10 predictions were the air-hexadecane partition coefficient and hydrogen bond acidity (Abraham parameters L and A) of the MPs and the dissolved organic carbon concentration of the GAC influent water. The model can be used to rapidly estimate the GAC bed life, select effective GAC products for a given treatment scenario, and explore the suitability of GAC treatment for remediating emerging MPs.