Independent External Dataset Research Articles

Neuroticism polygenic risk predicts conversion from mild cognitive impairment to Alzheimer's disease by impairing inferior parietal surface area.

The high prevalence of conversion from amnestic mild cognitive impairment (aMCI) to Alzheimer's disease (AD) makes early prevention of AD extremely critical. Neuroticism, a heritable personality trait associated with mental health, has been considered a risk factor for conversion from aMCI to AD. However, whether the neuroticism genetic risk could predict the conversion of aMCI and its underlying neural mechanisms is unclear. Neuroticism polygenic risk score (N-PRS) was calculated in 278 aMCI patients with qualified genomic and neuroimaging data from ADNI. After 1-year follow-up, N-PRS in patients of aMCI-converted group was significantly greater than those in aMCI-stable group. Logistic and Cox survival regression revealed that N-PRS could significantly predict the early-stage conversion risk from aMCI to AD. These results were well replicated in an internal dataset and an independent external dataset of 933 aMCI patients from the UK Biobank. One sample Mendelian randomization analyses confirmed a potentially causal association from higher N-PRS to lower inferior parietal surface area to higher conversion risk of aMCI patients. These analyses indicated that neuroticism genetic risk may increase the conversion risk from aMCI to AD by impairing the inferior parietal structure.

Revealing the impact of autophagy-related genes in rheumatoid arthritis: Insights from bioinformatics

BackgroundRheumatoid arthritis is a systemic inflammatory autoimmune disease that severely impacts physical and mental health. Autophagy is a cellular process involving the degradation of cellular components in lysosomes. However, from a bioinformatics perspective, autophagy-related genes have not been comprehensively elucidated in rheumatoid arthritis. MethodsIn this study, we performed differential analysis of autophagy-related genes in rheumatoid arthritis patients using the GSE93272 dataset from the Gene Expression Omnibus database. Marker genes were screened by least absolute shrinkage and selection operator. Based on marker genes, we used unsupervised cluster analysis to elaborate different autophagy clusters, and further identified modules strongly associated with rheumatoid arthritis by weighted gene co-expression network analysis. In addition, we constructed four machine learning models, random forest model, support vector machine model, generalized linear model and extreme gradient boosting based on marker genes, and based on the optimal machine learning model, a nomogram model was constructed for distinguishing between normal individuals and rheumatoid arthritis patients. Finally, five external independent rheumatoid arthritis datasets were used for the validation of our results. ResultsThe results showed that autophagy-related genes had significant expression differences between normal individuals and osteoarthritis patients. Through least absolute shrinkage and selection operator screening, we identified 31 marker genes and found that they exhibited significant synergistic or antagonistic effects in rheumatoid arthritis, and immune cell infiltration analysis revealed significant changes in immune cell abundance. Subsequently, we elaborated different autophagy clusters (cluster 1 and cluster 2) using unsupervised cluster analysis. Next, further by weighted gene co-expression network analysis, we identified a brown module strongly associated with rheumatoid arthritis. In addition, we constructed a nomogram model for five marker genes (CDKN2A, TP53, ATG16L2, FKBP1A, and GABARAPL1) based on a generalized linear model (area under the curve = 1.000), and the predictive efficiency and accuracy of this nomogram model were demonstrated in the calibration curves, the decision curves and the five external independent datasets were validated. ConclusionThis study identified marker autophagy-related genes in rheumatoid arthritis and analyzed their impact on the disease, providing new perspectives for understanding the role of autophagy-related genes in rheumatoid arthritis and providing new directions for its individualized treatment.

Artificial intelligence for ultrasound microflow imaging in breast cancer diagnosis.

To develop and evaluate artificial intelligence (AI) algorithms for ultrasound (US) microflow imaging (MFI) in breast cancer diagnosis. We retrospectively collected a dataset consisting of 516 breast lesions (364 benign and 152 malignant) in 471 women who underwent B-mode US and MFI. The internal dataset was split into training (n = 410) and test datasets (n = 106) for developing AI algorithms from deep convolutional neural networks from MFI. AI algorithms were trained to provide malignancy risk (0-100%). The developed AI algorithms were further validated with an independent external dataset of 264 lesions (229 benign and 35 malignant). The diagnostic performance of B-mode US, AI algorithms, or their combinations was evaluated by calculating the area under the receiver operating characteristic curve (AUROC). The AUROC of the developed three AI algorithms (0.955-0.966) was higher than that of B-mode US (0.842, P < 0.0001). The AUROC of the AI algorithms on the external validation dataset (0.892-0.920) was similar to that of the test dataset. Among the AI algorithms, no significant difference was found in all performance metrics combined with or without B-mode US. Combined B-mode US and AI algorithms had a higher AUROC (0.963-0.972) than that of B-mode US (P < 0.0001). Combining B-mode US and AI algorithms significantly decreased the false-positive rate of BI-RADS category 4A lesions from 87% to 13% (P < 0.0001). AI-based MFI diagnosed breast cancers with better performance than B-mode US, eliminating 74% of false-positive diagnoses in BI-RADS category 4A lesions.

A machine learning-based diagnostic model for myocardial infarction patients: Analysis of neutrophil extracellular traps-related genes and eQTL Mendelian randomization.

To identify neutrophil extracellular trap (NET)-associated gene features in the blood of patients with myocardial infarction (MI) using bioinformatics and machine learning, with the aim of exploring potential diagnostic utility in atherosclerosis. The datasets GSE66360 and GSE48060 were downloaded from the Gene Expression Omnibus (GEO) public database. GSE66360 was used as the training set, and GSE48060 was used as an independent validation set. Differential genes related to NETs were screened using R software. Machine learning was performed based on the differential expression of NET-related genes across different samples. The advantages and disadvantages of 4 machine learning algorithms (Random Forest [RF], Extreme Gradient Boosting [XGBoost, XGB], Generalized Linear Models [GLM], and Support Vector Machine-Recursive Feature Elimination [SVM-RFE]) were compared, and the optimal method was used to screen feature genes and construct diagnostic models, which were then validated in the external validation dataset. Correlations between feature genes and immune cells were analyzed, and samples were reclustered based on the expression of feature genes. Differences in downstream molecular mechanisms and immune responses were explored for different clusters. Weighted Gene Co-expression Network Analysis was performed on different clusters, and disease-related NET genes were extracted, followed by Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analysis. Finally, Mendelian randomization was employed to further investigate the causal relationship between the expression of model genes and the occurrence of MI. Forty-seven NET-related differential genes were obtained, and after comparing the 4 machine learning methods, support vector machine was used to screen ATG7, MMP9, interleukin 6 (IL6), DNASE1, and PDE4B as key genes for the construction of diagnostic models. The diagnostic value of the model was validated in an independent external validation dataset. These five genes showed strong correlations with neutrophils. Different sample clusters also demonstrated differential enrichment in pathways such as nitrogen metabolism, complement and coagulation cascades, cytokine-cytokine receptor interaction, renin-angiotensin system, and steroid biosynthesis. The Mendelian randomization results demonstrate a causal relationship between the expression of ATG7 and the incidence of myocardial infarction. The feature genes ATG7, MMP9, IL6, DNASE1, and PDE4B, identified using bioinformatics, may serve as potential diagnostic biomarkers and therapeutic targets for Myocardial infarction. Specifically, the expression of ATG7 could potentially be a significant factor in the occurrence of MI.

Abstract 886: Integrated deep learning model for predicting DNA methylation and tumor types from histopathology in central nervous system tumors

Abstract Precision in diagnosis of diverse central nervous system (CNS) tumor types is crucial for optimal patient treatment. DNA methylation profiles, which capture the methylation status of thousands of individual CpG sites, are data-driven means to enhance diagnostic accuracy, but this technique is expensive, time-consuming, and not yet routinely available. To address this, we developed DEPLOY, a deep-learning model that predicts 10 major categories of CNS tumors from histopathology. DEPLOY integrates three distinct components: the first classifies CNS tumors directly from histopathology slide images (‘direct model’); the second initially generates predictions for DNA methylation beta values, which are subsequently used for tumor classification (‘indirect model’); and the third classifies tumor types directly from routinely available patient demographics. First, we find that DEPLOY accurately predicted beta values from histopathology images, suggesting that application of inferred methylation data is a promising approach for deep learning histopathology-based classifiers. Using a 10-class model trained on an internal dataset, we apply this model on two independent external test datasets of 1,522 and 348 cases, achieving top-1 accuracies of 96% and 94%, respectively on samples which are predicted with high confidence. Taken together, DEPLOY could assist pathologists in diagnosing CNS tumors, in an equitable very low-cost manner, within a clinically relevant short time frame. Citation Format: Eldad David Shulman, Danh-Tai Hoang, Rust Turakulov, Zied Abdullaev, Omkar Singh, Emma M. Campagnolo, Eric A. Stone, MacLean P. Nasrallah, Eytan Ruppin, Kenneth Aldape. Integrated deep learning model for predicting DNA methylation and tumor types from histopathology in central nervous system tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 886.

Re-tear after arthroscopic rotator cuff repair can be predicted using deep learning algorithm.

The application of artificial intelligence technology in the medical field has become increasingly prevalent, yet there remains significant room for exploration in its deep implementation. Within the field of orthopedics, which integrates closely with AI due to its extensive data requirements, rotator cuff injuries are a commonly encountered condition in joint motion. One of the most severe complications following rotator cuff repair surgery is the recurrence of tears, which has a significant impact on both patients and healthcare professionals. To address this issue, we utilized the innovative EV-GCN algorithm to train a predictive model. We collected medical records of 1,631 patients who underwent rotator cuff repair surgery at a single center over a span of 5 years. In the end, our model successfully predicted postoperative re-tear before the surgery using 62 preoperative variables with an accuracy of 96.93%, and achieved an accuracy of 79.55% on an independent external dataset of 518 cases from other centers. This model outperforms human doctors in predicting outcomes with high accuracy. Through this methodology and research, our aim is to utilize preoperative prediction models to assist in making informed medical decisions during and after surgery, leading to improved treatment effectiveness. This research method and strategy can be applied to other medical fields, and the research findings can assist in making healthcare decisions.

An assessment system for clinical and biological interpretability in ulcerative colitis.

Ulcerative colitis (UC) is a serious inflammatory bowel disease (IBD) with high morbidity and mortality worldwide. As the traditional diagnostic techniques have various limitations in the practice and diagnosis of early ulcerative colitis, it is necessary to develop new diagnostic models from molecular biology to supplement the existing methods. In this study, we developed a machine learning-based synthesis to construct an artificial intelligence diagnostic model for ulcerative colitis, and the correctness of the model is verified using an external independent dataset. According to the significantly expressed genes related to the occurrence of UC in the model, an unsupervised quantitative ulcerative colitis related score (UCRScore) based on principal coordinate analysis was established. The UCRScore is not only highly generalizable across UC bulk cohorts at different stages, but also highly generalizable across single-cell datasets, with the same effect in terms of cell numbers, activation pathways and mechanisms. As an important role of screening genes in disease occurrence, based on connectivity map analysis, 5 potential targeting molecular compounds were identified, which can be used as an additional supplement to the therapeutic of UC. Overall, this study provides a potential tool for differential diagnosis and assessment of bio-pathological changes in UC at the macroscopic level, providing an opportunity to optimize the diagnosis and treatment of UC.

Detection of seizures with ictal tachycardia, using heart rate variability and patient adaptive logistic regression machine learning methods: A hospital-based validation study.

Automated seizure detection of focal epileptic seizures is needed for objective seizure quantification to optimize the treatment of patients with epilepsy. Heart rate variability (HRV)-based seizure detection using patient-adaptive threshold with logistic regression machine learning (LRML) methods has presented promising performance in a study with a Danish patient cohort. The objective of this study was to assess the generalizability of the novel LRML seizure detection algorithm by validating it in a dataset recorded from long-term video-EEG monitoring (LTM) in a Brazilian patient cohort. Ictal and inter-ictal ECG-data epochs recorded during LTM were analyzed retrospectively. Thirty-four patients had 107 seizures (79 focal, 28 generalized tonic-clonic [GTC] including focal-to-bilateral-tonic-clonic seizures) eligible for analysis, with a total of 185.5 h recording. Because HRV-based seizure detection is only suitable in patients with marked ictal autonomic change, patients with >50 beats/min change in heart rate during seizures were selected as responders. The patient-adaptive LRML seizure detection algorithm was applied to all elected ECG data, and results were computed separately for responders and non-responders. The patient-adaptive LRML seizure detection algorithm yielded a sensitivity of 84.8% (95% CI: 75.6-93.9) with a false alarm rate of .25/24 h in the responder group (22 patients, 59 seizures). Twenty-five of the 26 GTC seizures were detected (96.2%), and 25 of the 33 focal seizures without bilateral convulsions were detected (75.8%). The study confirms in a new, independent external dataset the good performance of seizure detection from a previous study and suggests that the method is generalizable. This method seems useful for detecting both generalized and focal epileptic seizures. The algorithm can be embedded in a wearable seizure detection system to alert patients and caregivers of seizures and generate objective seizure counts helping to optimize the treatment of the patients.

Auto-segmentation of pelvic organs at risk on 0.35T MRI using 2D and 3D Generative Adversarial Network models

PurposeManual recontouring of targets and Organs At Risk (OARs) is a time-consuming and operator-dependent task. We explored the potential of Generative Adversarial Networks (GAN) to auto-segment the rectum, bladder and femoral heads on 0.35T MRIs to accelerate the online MRI-guided-Radiotherapy (MRIgRT) workflow. Methods3D planning MRIs from 60 prostate cancer patients treated with 0.35T MR-Linac were collected. A 3D GAN architecture and its equivalent 2D version were trained, validated and tested on 40, 10 and 10 patients respectively. The volumetric Dice Similarity Coefficient (DSC) and 95th percentile Hausdorff Distance (HD95th) were computed against expert drawn ground-truth delineations. The networks were also validated on an independent external dataset of 16 patients. ResultsIn the internal test set, the 3D and 2D GANs showed DSC/HD95th of 0.83/9.72 mm and 0.81/10.65 mm for the rectum, 0.92/5.91 mm and 0.85/15.72 mm for the bladder, and 0.94/3.62 mm and 0.90/9.49 mm for the femoral heads. In the external test set, the performance was 0.74/31.13 mm and 0.72/25.07 mm for the rectum, 0.92/9.46 mm and 0.88/11.28 mm for the bladder, and 0.89/7.00 mm and 0.88/10.06 mm for the femoral heads. The 3D and 2D GANs required on average 1.44 s and 6.59 s respectively to generate the OARs’ volumetric segmentation for a single patient. ConclusionsThe proposed 3D GAN auto-segments pelvic OARs with high accuracy on 0.35T, in both the internal and the external test sets, outperforming its 2D equivalent in both segmentation robustness and volume generation time.

Identifying molecular subtypes and tumor microenvironment infiltration signatures in kidney renal clear cell carcinoma based on stemness-associated disulfidptosis genes by integrating machine learning, single-cell analyses and experimental validation

Clear cell renal cell carcinoma (ccRCC) is an aggressive malignant tumor. Disulfidptosis is a new programmed cell death mechanism, which is characterized by the abnormal accumulation of intracellular disulfides that are highly toxic to cells. However, the contribution of disulfidptosis to ccRCC progression has not been fully clarified. In this study, two different molecular subtypes related to disulfidptosis were identified in ccRCC patients by the non-negative matrix factorization (NMF) algorithm. The cluster 1 was characterized by a worse prognosis and higher mRNAsi levels. Then, difference analysis and weighted gene co-expression network analysis (WGCNA) were conducted to search modular genes that are highly associated with tumor stemness and tumor microenvironment. Subsequently, a SADG signature containing nine genes was constructed stepwise by WGCNA and least absolute shrinkage and selection operator (LASSO) Cox regression analysis. The high-risk score group had a worse outcome, and immune regulation and metabolic signatures might be responsible for cancer progression in the high-risk group. After that, a predictive nomogram was constructed, and the predicting power of the risk model was verified using inter and three independent external validation datasets. Nine SADGs were shown to significantly correlate with immune infiltration, tumor mutation burden (TMB), microsatellite instability (MSI) and immune checkpoint. In addition, based on the single-cell RNA sequencing dataset (GSE139555), the distribution and expression of nine hub genes in various types of immune cells were analyzed. Finally, the expression level of the nine genes was verified in clinical samples by qRT-PCR.

Development of a generative deep learning model to improve epiretinal membrane detection in fundus photography

BackgroundThe epiretinal membrane (ERM) is a common retinal disorder characterized by abnormal fibrocellular tissue at the vitreomacular interface. Most patients with ERM are asymptomatic at early stages. Therefore, screening for ERM will become increasingly important. Despite the high prevalence of ERM, few deep learning studies have investigated ERM detection in the color fundus photography (CFP) domain. In this study, we built a generative model to enhance ERM detection performance in the CFP.MethodsThis deep learning study retrospectively collected 302 ERM and 1,250 healthy CFP data points from a healthcare center. The generative model using StyleGAN2 was trained using single-center data. EfficientNetB0 with StyleGAN2-based augmentation was validated using independent internal single-center data and external datasets. We randomly assigned healthcare center data to the development (80%) and internal validation (20%) datasets. Data from two publicly accessible sources were used as external validation datasets.ResultsStyleGAN2 facilitated realistic CFP synthesis with the characteristic cellophane reflex features of the ERM. The proposed method with StyleGAN2-based augmentation outperformed the typical transfer learning without a generative adversarial network. The proposed model achieved an area under the receiver operating characteristic (AUC) curve of 0.926 for internal validation. AUCs of 0.951 and 0.914 were obtained for the two external validation datasets. Compared with the deep learning model without augmentation, StyleGAN2-based augmentation improved the detection performance and contributed to the focus on the location of the ERM.ConclusionsWe proposed an ERM detection model by synthesizing realistic CFP images with the pathological features of ERM through generative deep learning. We believe that our deep learning framework will help achieve a more accurate detection of ERM in a limited data setting.

Integrated multi-omic analysis and experiment reveals the role of endoplasmic reticulum stress in lung adenocarcinoma

BackgroundLung cancer is a highly prevalent malignancy worldwide and is associated with high mortality rates. While the involvement of endoplasmic reticulum (ER) stress in the development of lung adenocarcinoma (LUAD) has been established, the underlying mechanism remains unclear.MethodsIn this study, we utilized data from The Cancer Genome Atlas (TCGA) to identify differentially expressed endoplasmic reticulum stress-related genes (ERSRGs) between LUAD and normal tissues. We performed various bioinformatics analyses to investigate the biological functions of these ERSRGs. Using LASSO analysis and multivariate stepwise regression, we constructed a novel prognostic model based on the ERSRGs. We further validated the performance of the model using two independent datasets from the Gene Expression Omnibus (GEO). Additionally, we conducted functional enrichment analysis, immune checkpoint analysis, and immune infiltration analysis and drug sensitivity analysis of LUAD patients to explore the potential biological function of the model. Furthermore, we conducted a battery of experiments to verify the expression of ERSRGs in a real-world cohort.ResultsWe identified 106 ERSRGs associated with LUAD, which allowed us to classify LUAD patients into two subtypes based on gene expression differences. Using six prognostic genes (NUPR1, RHBDD2, VCP, BAK1, EIF2AK3, MBTPS2), we constructed a prognostic model that exhibited excellent predictive performance in the training dataset and was successfully validated in two independent external datasets. The risk score derived from this model emerged as an independent prognostic factor for LUAD. Confirmation of the linkage between this risk model and immune infiltration was affirmed through the utilization of Gene Set Enrichment Analysis (GSEA), Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses. The q-PCR results verified significant differences in the expression of prognostic genes between cancer and paracancer tissues. Notably, the protein expression of NUPR1, as determined by immunohistochemistry (IHC), exhibited an opposite pattern compared to the mRNA expression patterns.ConclusionThis study establishes a novel prognostic model for LUAD based on six ER stress-related genes, facilitating the prediction of LUAD prognosis. Additionally, NUPR1 was identified as a potential regulator of stress in LUAD.

Strengthening deep-learning models for intracranial hemorrhage detection: strongly annotated computed tomography images and model ensembles.

Multiple attempts at intracranial hemorrhage (ICH) detection using deep-learning techniques have been plagued by clinical failures. We aimed to compare the performance of a deep-learning algorithm for ICH detection trained on strongly and weakly annotated datasets, and to assess whether a weighted ensemble model that integrates separate models trained using datasets with different ICH improves performance. We used brain CT scans from the Radiological Society of North America (27,861 CT scans, 3,528 ICHs) and AI-Hub (53,045 CT scans, 7,013 ICHs) for training. DenseNet121, InceptionResNetV2, MobileNetV2, and VGG19 were trained on strongly and weakly annotated datasets and compared using independent external test datasets. We then developed a weighted ensemble model combining separate models trained on all ICH, subdural hemorrhage (SDH), subarachnoid hemorrhage (SAH), and small-lesion ICH cases. The final weighted ensemble model was compared to four well-known deep-learning models. After external testing, six neurologists reviewed 91 ICH cases difficult for AI and humans. InceptionResNetV2, MobileNetV2, and VGG19 models outperformed when trained on strongly annotated datasets. A weighted ensemble model combining models trained on SDH, SAH, and small-lesion ICH had a higher AUC, compared with a model trained on all ICH cases only. This model outperformed four deep-learning models (AUC [95% C.I.]: Ensemble model, 0.953[0.938-0.965]; InceptionResNetV2, 0.852[0.828-0.873]; DenseNet121, 0.875[0.852-0.895]; VGG19, 0.796[0.770-0.821]; MobileNetV2, 0.650[0.620-0.680]; p < 0.0001). In addition, the case review showed that a better understanding and management of difficult cases may facilitate clinical use of ICH detection algorithms. We propose a weighted ensemble model for ICH detection, trained on large-scale, strongly annotated CT scans, as no model can capture all aspects of complex tasks.

Machine learning based outcome prediction of microsurgically treated unruptured intracranial aneurysms

Machine learning (ML) has revolutionized data processing in recent years. This study presents the results of the first prediction models based on a long-term monocentric data registry of patients with microsurgically treated unruptured intracranial aneurysms (UIAs) using a temporal train-test split. Temporal train-test splits allow to simulate prospective validation, and therefore provide more accurate estimations of a model’s predictive quality when applied to future patients. ML models for the prediction of the Glasgow outcome scale, modified Rankin Scale (mRS), and new transient or permanent neurological deficits (output variables) were created from all UIA patients that underwent microsurgery at the Kepler University Hospital Linz (Austria) between 2002 and 2020 (n = 466), based on 18 patient- and 10 aneurysm-specific preoperative parameters (input variables). Train-test splitting was performed with a temporal split for outcome prediction in microsurgical therapy of UIA. Moreover, an external validation was conducted on an independent external data set (n = 256) of the Department of Neurosurgery, University Medical Centre Hamburg-Eppendorf. In total, 722 aneurysms were included in this study. A postoperative mRS > 2 was best predicted by a quadratic discriminant analysis (QDA) estimator in the internal test set, with an area under the receiver operating characteristic curve (ROC-AUC) of 0.87 ± 0.03 and a sensitivity and specificity of 0.83 ± 0.08 and 0.71 ± 0.07, respectively. A Multilayer Perceptron predicted the post- to preoperative mRS difference > 1 with a ROC-AUC of 0.70 ± 0.02 and a sensitivity and specificity of 0.74 ± 0.07 and 0.50 ± 0.04, respectively. The QDA was the best model for predicting a permanent new neurological deficit with a ROC-AUC of 0.71 ± 0.04 and a sensitivity and specificity of 0.65 ± 0.24 and 0.60 ± 0.12, respectively. Furthermore, these models performed significantly better than the classic logistic regression models (p < 0.0001). The present results showed good performance in predicting functional and clinical outcomes after microsurgical therapy of UIAs in the internal data set, especially for the main outcome parameters, mRS and permanent neurological deficit. The external validation showed poor discrimination with ROC-AUC values of 0.61, 0.53 and 0.58 respectively for predicting a postoperative mRS > 2, a pre- and postoperative difference in mRS > 1 point and a GOS < 5. Therefore, generalizability of the models could not be demonstrated in the external validation. A SHapley Additive exPlanations (SHAP) analysis revealed that this is due to the most important features being distributed quite differently in the internal and external data sets. The implementation of newly available data and the merging of larger databases to form more broad-based predictive models is imperative in the future.

10024-CO-1 A MULTICENTER PERFORMANCE COMPARISON STUDY OF AN IDH STATUS PREDICTION MODEL USING THE VISUAL TRANSFORMER MODEL

Abstract PURPOSE To compare the performance and explainability of the visual transformer (ViT) and convolutional neural network (CNN) architectures in predicting genomic mutations from brain MRI. METHODS The performances of the ViT and CNN classification models in predicting the IDH mutation status of gliomas were compared. The two models were fine-tuned on the TCIA dataset. The fine-tuned models were evaluated on the TCIA dataset and an external independent dataset, namely the Japanese Cohort (JC) dataset. To evaluate their explanatory power, the gradient-weighted class activation mapping (Grad-CAM) visualization of the CNNs model and attention map visualization of the ViT model were compared. RESULTS The visual transformer model consistently outperforms the convolutional neural network on both the TCIA and JC datasets (p-value = 0.021, p-value &amp;lt; 0.001, statistically different). The attention map of the ViT model accurately highlighted the tumor, and the Grad-CAM of the CNN model sometimes highlighted non-tumor areas. CONCLUSION The ViT model was more robust against differences in the image domain. The ViT model's attention map had superior explainability.

A generalizable and robust deep learning method for atrial fibrillation detection from long-term electrocardiogram

The reliable detection of atrial fibrillation (AF) is important for diagnosing the disease, tracking its progression, and developing individualized care strategies. However, models based on limited data are prone to data dependency due to differences in data feature distribution, which generally degrades their performance on unseen external datasets. In this work, we propose a multi-input fusion AF detection network (MIF-AFNet), which cascades residual convolutional neural networks and bidirectional long short-term memory networks to capture the deep features of electrocardiogram (ECG) and RR intervals (RRIs), respectively. Additionally, the ECG signals use a data augmentation method to alleviate the morphological imbalance. MIF-AFNet learns a robust feature representation for accurate AF detection by fusing the available information from RRIs and ECG. The proposed method was developed and evaluated using 5 long-term ECG datasets (CPSC2021, AFDB, LTAF, MITDB, and NSRDB) from PhysioNet. The subject-wise five-fold cross-validation was performed on CPSC2021, and the proposed method achieved an AF detection accuracy of 98.63%. The generalization performance is further evaluated on four external independent datasets (AFDB, LTAF, MITDB, and NSRDB), achieving accuracies of 98.63%, 97.04%, 98.07%, and 100%, respectively. The results show that the proposed method can accurately detect AF from long-term ECG recordings. In addition, the low complexity of the model makes it less demanding on computing resources. Therefore, it has the potential to improve the automatic diagnosis and management of AF in wearable device-based long-term home monitoring.

Identification of potential immune-related hub genes in Parkinson's disease based on machine learning and development and validation of a diagnostic classification model.

Parkinson's disease is the second most common neurodegenerative disease in the world. However, current diagnostic methods are still limited, and available treatments can only mitigate the symptoms of the disease, not reverse it at the root. The immune function has been identified as playing a role in PD, but the exact mechanism is unknown. This study aimed to search for potential immune-related hub genes in Parkinson's disease, find relevant immune infiltration patterns, and develop a categorical diagnostic model. We downloaded the GSE8397 dataset from the GEO database, which contains gene expression microarray data for 15 healthy human SN samples and 24 PD patient SN samples. Screening for PD-related DEGs using WGCNA and differential expression analysis. These PD-related DEGs were analyzed for GO and KEGG enrichment. Subsequently, hub genes (dld, dlk1, iars and ttd19) were screened by LASSO and mSVM-RFE machine learning algorithms. We used the ssGSEA algorithm to calculate and evaluate the differences in nigrostriatal immune cell types in the GSE8397 dataset. The association between dld, dlk1, iars and ttc19 and 28 immune cells was investigated. Using the GSEA and GSVA algorithms, we analyzed the biological functions associated with immune-related hub genes. Establishment of a ceRNA regulatory network for immune-related hub genes. Finally, a logistic regression model was used to develop a PD classification diagnostic model, and the accuracy of the model was verified in three independent data sets. The three independent datasets are GES49036 (containing 8 healthy human nigrostriatal tissue samples and 15 PD patient nigrostriatal tissue samples), GSE20292 (containing 18 healthy human nigrostriatal tissue samples and 11 PD patient nigrostriatal tissue samples) and GSE7621 (containing 9 healthy human nigrostriatal tissue samples and 16 PD patient nigrostriatal tissue samples). Ultimately, we screened for four immune-related Parkinson's disease hub genes. Among them, the AUC values of dlk1, dld and ttc19 in GSE8397 and three other independent external datasets were all greater than 0.7, indicating that these three genes have a certain level of accuracy. The iars gene had an AUC value greater than 0.7 in GES8397 and one independent external data while the AUC values in the other two independent external data sets ranged between 0.5 and 0.7. These results suggest that iars also has some research value. We successfully constructed a categorical diagnostic model based on these four immune-related Parkinson's disease hub genes, and the AUC values of the joint diagnostic model were greater than 0.9 in both GSE8397 and three independent external datasets. These results indicate that the categorical diagnostic model has a good ability to distinguish between healthy individuals and Parkinson's disease patients. In addition, ceRNA networks reveal complex regulatory relationships based on immune-related hub genes. In this study, four immune-related PD hub genes (dld, dlk1, iars and ttd19) were obtained. A reliable diagnostic model for PD classification was developed. This study provides algorithmic-level support to explore the immune-related mechanisms of PD and the prediction of immune-related drug targets.

Identification of CXCL16 as a diagnostic biomarker for obesity and intervertebral disc degeneration based on machine learning

Intervertebral disc degeneration (IDD) is the primary cause of neck and back pain. Obesity has been established as a significant risk factor for IDD. The objective of this study was to explore the molecular mechanisms affecting obesity and IDD by identifying the overlapping crosstalk genes associated with both conditions. The identification of specific diagnostic biomarkers for obesity and IDD would have crucial clinical implications. We obtained gene expression profiles of GSE70362 and GSE152991 from the Gene Expression Omnibus, followed by their analysis using two machine learning algorithms, least absolute shrinkage and selection operator and support vector machine-recursive feature elimination, which enabled the identification of C-X-C motif chemokine ligand 16 (CXCL16) as a shared diagnostic biomarker for obesity and IDD. Additionally, gene set variant analysis was used to explore the potential mechanism of CXCL16 in these diseases, and CXCL16 was found to affect IDD through its effect on fatty acid metabolism. Furthermore, correlation analysis between CXCL16 and immune cells demonstrated that CXCL16 negatively regulated T helper 17 cells to promote IDD. Finally, independent external datasets (GSE124272 and GSE59034) were used to verify the diagnostic efficacy of CXCL16. In conclusion, a common diagnostic biomarker for obesity and IDD, CXCL16, was identified using a machine learning algorithm. This study provides a new perspective for exploring the possible mechanisms by which obesity impacts the development of IDD.

Advancing Italian biomedical information extraction with transformers-based models: Methodological insights and multicenter practical application

The introduction of computerized medical records in hospitals has reduced burdensome activities like manual writing and information fetching. However, the data contained in medical records are still far underutilized, primarily because extracting data from unstructured textual medical records takes time and effort. Information Extraction, a subfield of Natural Language Processing, can help clinical practitioners overcome this limitation by using automated text-mining pipelines. In this work, we created the first Italian neuropsychiatric Named Entity Recognition dataset, PsyNIT, and used it to develop a Transformers-based model. Moreover, we collected and leveraged three external independent datasets to implement an effective multicenter model, with overall F1-score 84.77 %, Precision 83.16 %, Recall 86.44 %. The lessons learned are: (i) the crucial role of a consistent annotation process and (ii) a fine-tuning strategy that combines classical methods with a “low-resource” approach. This allowed us to establish methodological guidelines that pave the way for Natural Language Processing studies in less-resourced languages.

Value of Artificial Intelligence in Improving the Accuracy of Diagnosing TI-RADS Category 4 Nodules.

Considerable heterogeneity is observed in the malignancy rates of thyroid nodules classified as category 4 according to the Thyroid Imaging Reporting and Data System (TI-RADS). This study was aimed at comparing the diagnostic performance of artificial intelligence algorithms and radiologists with different experience levels in distinguishing benign and malignant TI-RADS 4 (TR4) nodules. Between January 2019 and September 2022, 1117 TR4 nodules with well-defined pathological findings were collected for this retrospective study. An independent external data set of 125 TR4 nodules was incorporated for testing purposes. Traditional feature-based machine learning (ML) models, deep convolutional neural networks (DCNN) models and a fusion model that integrated the prediction outcomes from all models were used to classify benign and malignant TR4 nodules. A fivefold cross-validation approach was employed, and the diagnostic performance of each model and radiologists was compared. In the external test data set, the area under the receiver operating characteristic curve (AUROC) of the three DCNN-based secondary transfer learning models-InceptionV3, DenseNet121 and ResNet50-were 0.852, 0.837 and 0.856, respectively. These values were higher than those of the three traditional ML models-logistic regression, multilayer perceptron and random forest-at 0.782, 0.790, and 0.767, respectively, and higher than that of an experienced radiologist (0.815). The fusion diagnostic model we developed, with an AUROC of 0.880, was found to outperform the experienced radiologist in diagnosing TR4 nodules. The integration of artificial intelligence algorithms into medical imaging studies could improve the accuracy of identifying high-risk TR4 nodules pre-operatively and have significant clinical application potential.