AI-based Model Research Articles

Examination of AI Algorithms for Image and MRI-based Autism Detection

Precise identification of autism spectrum disorder (ASD) is a challenging task due to the heterogeneity of ASD. Early diagnosis and interventions have positive effects on treatment and later skills development. Hence, it is necessary to provide families and communities with the resources, training, and tools required to diagnose and help patients. Recent work has shown that artificial intelligence-based methods are suitable for the identification of ASD. AI-based tools can be good resources for parents for early detection of ASD in their kids. Even AI-based advanced tools are helpful for health workers and physicians to detect ASD. Facial images and MRI are the best sources to understand ASD symptoms, hence are input required in AI-based model training. The trained models are used for the classification of ASD patients and normal kids. The deep learning models are found to be very accurate in ASD detection. In this paper, we present a comprehensive study of AI techniques like machine learning, image processing, and deep learning, and their accuracy when these techniques are used on facial and MRI images of ASD and normally developed kids.

Validation and Retraining of the Stellae-123 Gene Expression Signature Improved Risk Stratification in Taiwanese Acute Myeloid Leukemia Patients

The European Leukemia Net (ELN) recommendations provide valuable guidance in treatment decisions of patients with acute myeloid leukemia (AML). However, the genetic complexity and heterogeneity of AML are not fully covered, despite the critical role of gene expression analysis in risk stratification. Addressing this gap, the Stellae-123 score, an AI-based model that captures gene expression patterns, has demonstrated promising results in predicting survival of AML patients in four western cohorts. This study aims to assess the applicability of Stellae-123 in a Taiwanese cohort. Our cohort included 233 de novo AML patients, who were diagnosed and treated at the National Taiwan University Hospital and received standard induction therapy with 7+3 chemotherapy (or 5+2 in elder fit patients). All patients had sufficient bone marrow samples for DNA and RNA sequencing at diagnosis. The median age of the cohort was 41 years. Among the patients, ELN-2022 risk groups were distributed as follows: favorable (53.4%), intermediate (27.6%), and adverse (19%). Appling the original Stellae-123 model, which derived gene expression signatures from the BeatAML cohort RNA-seq data, our patients were categorized into low-, intermediate-, and high-risk groups, each representing 33% of the cohort. Overall, there were no differences in clinical features among these risk groups. The Stellae-123 risk groups partially aligned with the ELN risk groups, with more ELN low-risk patients in the Stellae-123 low-risk group and a higher proportion of ELN-high risk patients in the Stellae-123 high-risk group. Intriguingly, aside from mutations listed in ELN-2022, the Stellae-123 high-risk group had significantly less NRAS mutations than others. Cumulative hazards predicted by Stellae-123 model yielded a c-index of 0.612 for overall survival (OS), and the three risk groups had significantly different OS (p<0.001). We next postulated that since Stellae-123 relies on molecular features derived from the leukemic cells, it might serve as a biomarker of biological risk to predict relapse-free survival (RFS). The c-index calculated by Stellae-123 to predict RFS was 0.607, suggesting a reasonable ability to discriminate between different risk levels for RFS. Recognizing the potential variation between the original training cohort and the Taiwanese population, we addressed this concern by retraining the Stellae-123 model using Taiwanese RNAseq data. As a result, the model retraining led to significant improvements in the cross-validated c-indexes, with values of 0.656 (OS) and 0.636 (RFS), leading to an even more robust risk stratification (p<0.001, Figure 1A). As we noticed overlapping grouping between Stellae-123 and ELN-2022 in both pretrained and retrained models, we explored if Stellae-123's prognostic power was independent of ELN risk status by conducting a multivariable analysis. Remarkably, in both the pretrained and retrained Stellae-123 models, whether calculated as continuous values or categorized into three groups, they consistently demonstrated discriminative power in predicting OS and RFS, irrespective of age and ELN-2022 risk, suggesting that Stellae-123 could offer additional valuable information in predicting patient outcomes. Considering recent data showing that incorporating age helped improve the prognostication performance of ELN-2022 and that age was an independent risk factor for inferior outcome in our cohort, we examined how age could complement current risk stratification. Serial model testing revealed significant improvement of the prognostic models when age and transcriptomic data, especially the locally retrained Stellae-123 model, were combined with ELN-2022, evident by drastically declined Akaike Information Criterion (AIC) values. Time-dependent ROC curve analysis confirmed the potential of incorporating Stellae-123 and age to complement current risk stratification system (Figure 1B). In summary, we showed that the Stellae-123 predicted both OS and relapse risk in AML patients from Taiwan. Furthermore, we demonstrated that model retraining could improve prognostication by adapting the Stellae-123 to the particularities of Taiwanese patients. Given the remarkable extrapolation of the signature, it's suggested gene expression profile be incorporated in the risk stratification of AML patients who are candidates to intensive therapy.

Explainable Deep Learning for Mental Health Detection from English and Arabic Social Media Posts

The research communities have recently begun exploring the detection of depression through social media, making it a relatively new development in this field. Some research has been done to identify depressive signs from English social media postings, while little work has been done for the Arabic posts. This paper proposes a BERT and Bi-LSTM pipeline for the identification of depressive signs from Arabic social media posts. A fine-tuned RoBERTa-based model is also proposed for English social media posts for depressive state identification. Along with the proposed model, seven conventional machine learning and eight deep learning models are also explored for the identification of depressive signs from Arabic and English social media posts. The performance of the proposed model is validated on two Arabic datasets and one English dataset. The proposed BERT and Bi-LSTM pipeline achieved state-of-the-art performance with an F 1 -score of 1.00 and 0.82 for two different Arabic datasets, whereas the proposed fine-tuned RoBERTa achieved a F 1 -score of 0.60 which is comparable in identifying depressive sign from English social media posts. The majority of the suggested deep learning models are end-to-end, which necessitates a greater explanation for their success. An explainable AI-based model may enhance decision-making, transparency, and interpretability. Therefore, this research identifies where the suggested system learned well and where it failed in recognition of the depression signs that can help in future developments in the field of depression detection.

Artificial intelligence-based identification of left ventricular systolic dysfunction from 12-lead electrocardiograms: External validation and advanced application of an existing model

Abstract Background The diagnostic application of artificial intelligence (AI)-based models to detect cardiovascular diseases from electrocardiograms (ECG) evolves and some promising results were reported. However, most published algorithms lack external validation. Purpose Aim of this study was to validate an existing, open-source algorithm for the detection of left ventricular systolic dysfunction (LVSD) from 12-lead ECGs. Methods Patients with digitalized data pairs of 12-lead ECGs and echocardiography (at intervals of ≤7 days) were retrospectively selected from the Heart Center Leipzig ECG and electronic medical records databases. A previously developed AI-based model was applied to ECGs and calculated probabilities for LVSD. Area under the receiver operating characteristic curve (AUROC) was computed overall and in cohorts stratified for baseline and ECG characteristics. Repeated echocardiography studies recorded ≥3 months after index diagnostics were used for follow-up (FU) analysis. Results At baseline, 42,291 ECG-echocardiography pairs were analyzed and AUROC for LVSD detection was 0.88. Corresponding sensitivity and specificity were 82% and 77% for the optimal LVSD-probability cutoff based on Youden’s J. AUROCs were lower in ECG-subgroups with tachycardia, atrial fibrillation and wide QRS complex. In patients without LVSD at baseline and available FU, model-generated high probability for LVSD was associated with a 4-fold increased risk of developing LVSD during FU. Conclusion We provide the external validation of an existing AI-based ECG-analyzing model for the detection of LVSD with excellent and robust performance metrics. The association of false positive LVSD screenings at baseline with a deterioration of ventricular function during FU is hypothesis-generating and deserves further research.

Evaluation of the compressive strength of polypropylene fiber reinforced high-strength concrete support with AI-based model

Evaluation of the compressive strength of polypropylene fiber reinforced high-strength concrete support with AI-based model

A personalized classification of behavioral severity of autism spectrum disorder using a comprehensive machine learning framework

Autism Spectrum Disorder (ASD) is characterized as a neurodevelopmental disorder with a heterogeneous nature, influenced by genetics and exhibiting diverse clinical presentations. In this study, we dissect Autism Spectrum Disorder (ASD) into its behavioral components, mirroring the diagnostic process used in clinical settings. Morphological features are extracted from magnetic resonance imaging (MRI) scans, found in the publicly available dataset ABIDE II, identifying the most discriminative features that differentiate ASD within various behavioral domains. Then, each subject is categorized as having severe, moderate, or mild ASD, or typical neurodevelopment (TD), based on the behavioral domains of the Social Responsiveness Scale (SRS). Through this study, multiple artificial intelligence (AI) models are utilized for feature selection and classifying each ASD severity and behavioural group. A multivariate feature selection algorithm, investigating four different classifiers with linear and non-linear hypotheses, is applied iteratively while shuffling the training-validation subjects to find the set of cortical regions with statistically significant association with ASD. A set of six classifiers are optimized and trained on the selected set of features using 5-fold cross-validation for the purpose of severity classification for each behavioural group. Our AI-based model achieved an average accuracy of 96%, computed as the mean accuracy across the top-performing AI models for feature selection and severity classification across the different behavioral groups. The proposed AI model has the ability to accurately differentiate between the functionalities of specific brain regions, such as the left and right caudal middle frontal regions. We propose an AI-based model that dissects ASD into behavioral components. For each behavioral component, the AI-based model is capable of identifying the brain regions which are associated with ASD as well as utilizing those regions for diagnosis. The proposed system can increase the speed and accuracy of the diagnostic process and result in improved outcomes for individuals with ASD, highlighting the potential of AI in this area.

Applying an artificial intelligence deep learning approach to routine dermatopathological diagnosis of basal cell carcinoma.

Institutes of dermatopathology are faced with considerable challenges including a continuously rising numbers of submitted specimens and a shortage of specialized health care practitioners. Basal cell carcinoma (BCC) is one of the most common tumors in the fair-skinned western population and represents a major part of samples submitted for histological evaluation. Digitalizing glass slides has enabled the application of artificial intelligence (AI)-based procedures. To date, these methods have found only limited application in routine diagnostics. The aim of this study was to establish an AI-based model for automated BCC detection. In three dermatopathological centers, daily routine practice BCC cases were digitalized. The diagnosis was made both conventionally by analog microscope and digitally through an AI-supported algorithm based on a U-Net architecture neural network. In routine practice, the model achieved a sensitivity of 98.23% (center 1) and a specificity of 98.51%. The model generalized successfully without additional training to samples from the other centers, achieving similarly high accuracies in BCC detection (sensitivities of 97.67% and 98.57% and specificities of 96.77% and 98.73% in centers 2 and 3, respectively). In addition, automated AI-based basal cell carcinoma subtyping and tumor thickness measurement were established. AI-based methods can detect BCC with high accuracy in a routine clinical setting and significantly support dermatopathological work.

AI-based hip prosthesis failure prediction through evolutional radiological indices.

This study aimed to develop artificial intelligence models for predicting hip implant failure from radiological features. Analyzing the evolution of the periprosthetic bone and implant's position throughout the entire follow-up period has shown the potential to be more relevant in outcome prediction than simply considering the latest radiographic images. Thus, we investigated an AI-based model employing a small set of evolutional parameters derived from conventional radiological features to predict hip prosthesis failure. One hundred sixty-nine radiological features were annotated from historical anteroposterior and lateral radiographs for 162 total hip arthroplasty patients, 32 of which later underwent implant failure. Linear regression on each patient's chronologically sorted radiological features was employed to derive 169 corresponding evolutional parameters per image. Three sets of machine learning predictors were developed: one employing the original features (standard model), one the evolutional ones (evolutional model), and the last their union (hybrid model). Each set included a model employing all the available features (full model) and a model employing the few most predictive ones according to Gini importance (minimal model). The evolutional and hybrid predictors resulted highly effective (area under the ROC curve (AUC) of full models = 0.94), outperforming the standard one, whose AUC was only 0.82. The minimal hybrid model, employing just four features, three of which evolutional, scored an AUC of 0.95, proving even more accurate than the full one, exploiting 173 features. This tool could be shaped to be either highly specific (sensitivity: 80%, specificity: 98.6%) or highly sensitive (sensitivity: 90%, specificity: 92.4%). The proposed predictor may represent a highly sensitive screening tool for clinicians, capable to predict THA failure with an advance between a few months and more than a year through only four radiological parameters, considering either their value at the latest visit or their evolution through time.

Urinary Tract Infection Bacteria Classification: Artificial Intelligence-based Medical Application

Urinary tract infection (UTI) is a type of health disorder, an infection in the urinary glands mainly caused by bacteria. Currently, conventional early detection methods that have been established involve rapid dipstick strip test and urine culture analysis, which have suboptimal accuracy and effectiveness. Several retrospective studies regarding UTI bacteria classification have shown promising results, but still have limitations regarding prediction accuracy and technical simplicity. This study aims to implement a method based on artificial intelligence (AI) in classifying images of bacteria that causes UTIs. Eight artificial intelligence methods based on deep neural networks were used in the study; the models were evaluated and compared based on the prediction's effectiveness and accuracy. This study also seeks to create the easiest method of classifying bacteria causing UTIs using a computer-based application with the best obtained AI-based model. The best training results using an intelligent approach placed DenseNet201 as the method with the highest accuracy (83.99%). Then, the output model was used as a knowledge reference for the designed computer-based application. Real-time prediction results will appear in the application window.

Feasibility of artificial intelligence-driven interfractional monitoring of organ changes by mega-voltage computed tomography in intensity-modulated radiotherapy of prostate cancer.

High-dose radiotherapy (RT) for localized prostate cancer requires careful consideration of target position changes and adjacent organs-at-risk (OARs), such as the rectum and bladder. Therefore, daily monitoring of target position and OAR changes is crucial in minimizing interfractional dosimetric uncertainties. For efficient monitoring of the internal condition of patients, we assessed the feasibility of an auto-segmentation of OARs on the daily acquired images, such as megavoltage computed tomography (MVCT), via a commercial artificial intelligence (AI)-based solution in this study. We collected MVCT images weekly during the entire course of RT for 100 prostate cancer patients treated with the helical TomoTherapy system. Based on the manually contoured body outline, the bladder including prostate area, and rectal balloon regions for the 100 MVCT images, we trained the commercially available fully convolutional (FC)-DenseNet model and tested its auto-contouring performance. Based on the optimally determined hyperparameters, the FC-DenseNet model successfully auto-contoured all regions of interest showing high dice similarity coefficient (DSC) over 0.8 and a small mean surface distance (MSD) within 1.43 mm in reference to the manually contoured data. With this well-trained AI model, we have efficiently monitored the patient's internal condition through six MVCT scans, analyzing DSC, MSD, centroid, and volume differences. We have verified the feasibility of utilizing a commercial AI-based model for auto-segmentation with low-quality daily MVCT images. In the future, we will establish a fast and accurate auto-segmentation and internal organ monitoring system for efficiently determining the time for adaptive replanning.

Using Artificial Intelligence-Based Approach for Detecting Insects-Induced Grain Damage

According to the Intergovernmental Panel on Climate Change, around 30% of global food production is wasted each year. Prevention strategies during post-harvest grain storage may reduce food waste and help distribute food surpluses more fairly globally, ultimately reducing global hunger. Insect damage is a major cause of post-harvest storage losses, resulting in both quality and nutritional loss to the grain. Recently, data-driven approaches have been used to enhance proper and efficient storage. The aim of this project is to investigate how artificial intelligence (AI) can be used to detect insect-induced grain damage during storage. This project used barley grain to test the experimental hypothesis that insect-related damage during grain storage can be detected by simultaneous imaging and AI-based analysis. An AI approach was developed using Python to build a prediction model. Data was collected by acquiring images of insect-induced damaged and undamaged grain. The AI-based model was highly (90%) accurate in identifying the damage caused by the insect based on the parameters defined in the algorithm. In summary, this innovative approach allowed us to identify grain damage, which, in the future, will help take necessary intervention(s) to prevent insect-induced grain damage and ultimately prevent loss during post-harvest storage.

Overcoming Nonlinear Dynamics in Diabetic Retinopathy Classification: A Robust AI-Based Model with Chaotic Swarm Intelligence Optimization and Recurrent Long Short-Term Memory

Diabetic retinopathy (DR), which is seen in approximately one-third of diabetes patients worldwide, leads to irreversible vision loss and even blindness if not diagnosed and treated in time. It is vital to limit the progression of DR disease in order to prevent the loss of vision in diabetic patients. It is therefore essential that DR disease is diagnosed at an early phase. Thanks to retinal screening at least twice a year, DR disease can be diagnosed in its early phases. However, due to the variations and complexity of DR, it is really difficult to determine the phase of DR disease in current clinical diagnoses. This paper presents a robust artificial intelligence (AI)-based model that can overcome nonlinear dynamics with low computational complexity and high classification accuracy using fundus images to determine the phase of DR disease. The proposed model consists of four stages, excluding the preprocessing stage. In the preprocessing stage, fractal analysis is performed to reveal the presence of chaos in the dataset consisting of 12,500 color fundus images. In the first stage, two-dimensional stationary wavelet transform (2D-SWT) is applied to the dataset consisting of color fundus images in order to prevent information loss in the images and to reveal their characteristic features. In the second stage, 96 features are extracted by applying statistical- and entropy-based feature functions to approximate, horizontal, vertical, and diagonal matrices of 2D-SWT. In the third stage, the features that keep the classifier performance high are selected by a chaotic-based wrapper approach consisting of the k-nearest neighbor (kNN) and chaotic particle swarm optimization algorithms (CPSO) to cope with both chaoticity and computational complexity in the fundus images. At the last stage, an AI-based classification model is created with the recurrent neural network-long short-term memory (RNN-LSTM) architecture by selecting the lowest number of feature sets that can keep the classification performance high. The performance of the DR disease classification model was tested on 2500 color fundus image data, which included five classes: no DR, mild non-proliferative DR (NPDR), moderate NPDR, severe NPDR, and proliferative DR (PDR). The robustness of the DR disease classification model was confirmed by the 10-fold cross-validation. In addition, the classification performance of the proposed model is compared with the support vector machine (SVM), which is one of the machine learning techniques. The results obtained show that the proposed model can overcome nonlinear dynamics in color fundus images with low computational complexity and is very effective and successful in precisely diagnosing all phases of DR disease.

Development of an artificial intelligence-based diagnostic system for the detection of abnormal colposcopic findings.

82 Background: When performing colposcopic examinations in patients with abnormal cervical cytology, sufficient training is required to detect cervical intraepithelial neoplasia (CIN) with (1) high diagnostic accuracy and (2) minimizing time and reducing tissue biopsies. The aim of this study was to develop an artificial intelligence (AI)-based system that replicates expert colposcopic examination techniques to (1) identify CIN lesions with high accuracy and to (2) guide tissue sampling locations with high diagnostic performance, independent of examiner skill. Methods: A retrospective analysis was performed using 8341 colposcopic videos from 2013 to 2019, consisting of seven early-stage cervical cancer cases, 203 CIN3 cases, 276 CIN2 cases, and 456 CIN1 cases. An AI-based lesion detection model was constructed by annotating abnormal colposcopic findings after acetic acid processing in cervical cancer and CIN3 cases whose histologic diagnoses were confirmed by biopsies. The developed lesion detection model was applied to CIN1 and CIN2 cases, and the diagnostic accuracies of the lesions were evaluated. The accuracies of the lesions of CIN1 and CIN2 were evaluated according to the lesion area (sensitivity, specificity, area under the curve (AUC)) and the number of lesions identified. Results: The AI-based model was trained on 60 cases of cervical cancer and CIN3 and validated on 150 cases. The model was able to identify severe lesions with high accuracy, with a sensitivity of 85%, a specificity of 73%, an AUC of 0.89 for lesion area, and an accuracy of 95% for the number of lesions identified. The model also predicted abnormal colposcopic findings in CIN1 and CIN2 cases with high accuracy for detection of lesion area (sensitivity: 87% and 86%, specificity: 70% and 67%, AUC: 0.81 and 0.81, respectively) and identification of the number of lesions (97% and 93%, respectively). Furthermore, a heatmap display based on lesion prediction allowed visualization of the area with the highest acetic acid intensity corresponding to the actual biopsy locations. Conclusions: We have developed an AI-based diagnostic system for colposcopy that can identify CIN lesions with high accuracy and suggest appropriate biopsy sites. We are currently establishing an AI-based application for the visualization of abnormal lesions in colposcopy developed from the model, and anticipating its implementation in the clinical practice of colposcopic examinations.

Deep learning-based characterization of the drug tolerant persister cell state in lung cancer.

141 Background: Lung cancer is the leading cause of cancer-related lethality globally. Targeted therapies improve the clinical outcome of cancer treatment; however, a subpopulation of cancer cells survive during initial therapy and evolve into drug tolerant persister cells (DTPs) that maintain a residual disease reservoir. Residual disease preludes acquired resistance and tumor progression; therefore, identifying and eliminating DTPs could benefit future treatment paradigms. We have shown that Hippo pathway effector YAP1 (Yes Associated Protein-1) is activated in oncogene-driven lung cancers when cancer cells are exposed to various targeted therapies, such as EGFR, ALK, and RAS inhibitors. YAP1 transcriptional activation during targeted therapy is characterized by its increased nuclear localization and interaction with transcription factors. This activation promotes the expression of genes involved in cell survival, cellular plasticity, and metabolic reprogramming at the residual disease state. We hypothesized that detection of intrinsic or acquired persister cells may aid the development of optimized treatment regimens. In this study, we are building image-based deep learning models to identify YAP1 activation-mediated DTPs from histologically stained slides. Methods: H&E or YAP-stained immunohistochemistry (IHC) images from clinical lung cancer and patient-derived tumor xenograft samples were collected throughout targeted therapy and were annotated with semi-automation using high performance computing clusters. A deep learning model (U-Net algorithm) was used for image segmentation, training, validation, and testing. Results: 1638 images were annotated and over 80,000 patches from these images for YAP positive cells (or regions of interest) comprised the training dataset with semi-supervised automation. Subsequently, we built a customized deep-learning model to detect YAP-mediated DTP cell states from whole histopathological image slides. The deep learning-based model achieved excellent accuracy of 0.8238 and 0.9091 in training, and 0.8040 and 0.8949 in validation datasets for two different annotations, respectively. For a test dataset, the model obtained 0.81 and 0.902 accuracy for the two annotations, respectively. Conclusions: We have constructed a deep learning convolutional neural network model to infer the presence of the YAP1 activation-mediated drug tolerant persister cell state prior to or during targeted treatment of lung cancer. Implementing our AI-based model into routine lung cancer care in the future could identify patient subpopulations with YAP1 activated tumors who would most benefit from receiving YAP1-targeted small molecule inhibitors.

AI-based Virtual Synthesis of Methionine PET from Contrast-enhanced MRI: Development and External Validation Study.

Background Carbon 11 (11C)-methionine is a useful PET radiotracer for the management of patients with glioma, but radiation exposure and lack of molecular imaging facilities limit its use. Purpose To generate synthetic methionine PET images from contrast-enhanced (CE) MRI through an artificial intelligence (AI)-based image-to-image translation model and to compare its performance for grading and prognosis of gliomas with that of real PET. Materials and Methods An AI-based model to generate synthetic methionine PET images from CE MRI was developed and validated from patients who underwent both methionine PET and CE MRI at a university hospital from January 2007 to December 2018 (institutional data set). Pearson correlation coefficients for the maximum and mean tumor to background ratio (TBRmax and TBRmean, respectively) of methionine uptake and the lesion volume between synthetic and real PET were calculated. Two additional open-source glioma databases of preoperative CE MRI without methionine PET were used as the external test set. Using the TBRs, the area under the receiver operating characteristic curve (AUC) for classifying high-grade and low-grade gliomas and overall survival were evaluated. Results The institutional data set included 362 patients (mean age, 49 years ± 19 [SD]; 195 female, 167 male; training, n = 294; validation, n = 34; test, n = 34). In the internal test set, Pearson correlation coefficients were 0.68 (95% CI: 0.47, 0.81), 0.76 (95% CI: 0.59, 0.86), and 0.92 (95% CI: 0.85, 0.95) for TBRmax, TBRmean, and lesion volume, respectively. The external test set included 344 patients with gliomas (mean age, 53 years ± 15; 192 male, 152 female; high grade, n = 269). The AUC for TBRmax was 0.81 (95% CI: 0.75, 0.86) and the overall survival analysis showed a significant difference between the high (2-year survival rate, 27%) and low (2-year survival rate, 71%; P < .001) TBRmax groups. Conclusion The AI-based model-generated synthetic methionine PET images strongly correlated with real PET images and showed good performance for glioma grading and prognostication. Published under a CC BY 4.0 license. Supplemental material is available for this article.

An AI-Based Image Quality Control Framework for Knee Radiographs

Image quality control (QC) is crucial for the accurate diagnosis of knee diseases using radiographs. However, the manual QC process is subjective, labor intensive, and time-consuming. In this study, we aimed to develop an artificial intelligence (AI) model to automate the QC procedure typically performed by clinicians. We proposed an AI-based fully automatic QC model for knee radiographs using high-resolution net (HR-Net) to identify predefined key points in images. We then performed geometric calculations to transform the identified key points into three QC criteria, namely, anteroposterior (AP)/lateral (LAT) overlap ratios and LAT flexion angle. The proposed model was trained and validated using 2212 knee plain radiographs from 1208 patients and an additional 1572 knee radiographs from 753 patients collected from six external centers for further external validation. For the internal validation cohort, the proposed AI model and clinicians showed high intraclass consistency coefficients (ICCs) for AP/LAT fibular head overlap and LAT knee flexion angle of 0.952, 0.895, and 0.993, respectively. For the external validation cohort, the ICCs were also high, with values of 0.934, 0.856, and 0.991, respectively. There were no significant differences between the AI model and clinicians in any of the three QC criteria, and the AI model required significantly less measurement time than clinicians. The experimental results demonstrated that the AI model performed comparably to clinicians and required less time. Therefore, the proposed AI-based model has great potential as a convenient tool for clinical practice by automating the QC procedure for knee radiographs.

MOSUNETUZUMAB DEMONSTRATES DURABLE RESPONSES IN PATIENTS WITH RELAPSED/REFRACTORY FOLLICULAR LYMPHOMA AND ≥2 PRIOR THERAPIES: UPDATED ANALYSIS OF A PIVOTAL PHASE II STUDY

MOSUNETUZUMAB DEMONSTRATES DURABLE RESPONSES IN PATIENTS WITH RELAPSED/REFRACTORY FOLLICULAR LYMPHOMA AND ≥2 PRIOR THERAPIES: UPDATED ANALYSIS OF A PIVOTAL PHASE II STUDY

Explainable Paroxysmal Atrial Fibrillation Diagnosis Using Electrocardiogram with Artificial Intelligence

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Research Foundation of Korea (NRF) funded by the Korea government (MSIT) Background Paroxysmal atrial fibrillation (PAF) is common among patients with cryptogenic stroke or transient ischemic attack, and has a silent nature. Therefore, robust and reliable early detection of atrial fibrillation would be valuable for managing cardiovascular comorbidities. Therefore, robust and reliable early detection of atrial fibrillation would be valuable for managing cardiovascular comorbidities. Despite many efforts to predict or detect atrial fibrillation, an adequate explanation for estimation remains problematic. Method We present a statistical prediction model employing a deep-learning approach to identify a PAF from normal electrocardiogram (ECG), and interpret their causes. Based on 552,372 12-lead ECG data of 318,321 patients, we show that an AI-based model can capture the features of atrial fibrillation during normal sinus rhythm using ECG and additional characteristics. As a result of applying our labelling method, only 26,541 ECGs (4.8%) were annotated with PAF. We also applied Layer-wise Relevance Propagation(LRP) algorithm to investigate the regions of the ECG that primarily affect the discrimination of PAF. Results The performance of our model was evaluated as the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. The AUC for predicting PAF was 0.910. The predictive performance according to the class was 0.837 in the negative prediction and 0.705 in the positive prediction. The sensitivity analysis results showed that the difference in PAF prediction information for each lead was not significant. Among the features obtained from ECG, the PR interval and QT interval were significant explaining factors for the prediction of PAF, and this fact was confirmed by visualizing the ECG data using the LRP algorithm. Conclusion We expect that the discrimination of undetected atrial fibrillation can be used as a pre-emptive assistive tool to provide probabilistic predictions for the screening of PAF and could be enables insights into the cardiac dysfunction.

Analysis of long-term cardiovascular outcomes of COVID-19. Artificial Intelligence model for risk evaluation

Abstract Funding Acknowledgements Type of funding sources: None. Background The COVID-19 pandemic has caused a serious worldwide health crisis with wide-reaching consequences. According to current studies, COVID-19 may contribute to the emergence of various problems, including cardiovascular disorders [1-3]. However, there are currently no strategies for predicting the cardiovascular post-acute COVID-19 consequences, nor is there any scientific examination of these effects [3]. This knowledge gap must be filled in order to improve post-acute COVID-19 care [4]. Purpose To conduct a 12-month follow-up longitudinal observation study to explore potential long-term cardiovascular outcomes of COVID-19 and develop an AI-based model to predict them. Methods We included 328 COVID-19 patients who were admitted to the hospital between February and April 2021. The average age of the studied population was 56.1, while 51% of the population was male. Patients with severe comorbidities, major adverse cardiovascular events (MACE) in anamnesis, and intra-hospital mortality were not included. After discharge, patients were monitored for 12 months. The data were randomly divided into derivation (n=201), validation (n=81), and external cohort (n=47). The obtained data include clinical data, results of laboratory and instrumental investigations, and medical records during the 1-year period. Our AI model is a convolutional neural network (CNN) consisting of sigmoid-activation-function neurons, trained by using outcome results, received in follow-ups, and evaluated on the external cohort. Results Due to the analysis of clinical data, higher levels of C-reactive protein, D-dimer, neutrophil/lymphocyte ratio, and lower levels of thyrotropin were associated with a higher 12-month risk of cardiovascular consequences. The longitudinal study showed that 16.4% of patients experienced MACE (8.5% - myocardial infarction, 5.2% - cerebrovascular disorders, 2.7% - pulmonary embolism) leading to cardiovascular death in 2.4% of the studied population. The total count of cardiovascular complications was 25.9% including first-detected hypertension, arrhythmias, heart failure etc. The derivation and validation datasets were used for the training of the CNN. The ROC analysis indicated that our AI model had a high predictive value for cardiovascular events during follow-up (AUC=0.891, 95% CI=0.869–0.913, p&amp;lt;0.001, sensitivity 92.4%, specificity 90.1%). The external cohort data analysis showed accurate results in 41/47 patients (87.2% accuracy), which significantly outperformed any other known methods of cardiovascular outcomes prediction during 12 months after COVID-19. Conclusions Cardiovascular health in COVID-19 survivors needs special attention since COVID-19 is linked to an elevated risk of post-acute cardiovascular consequences. Long-term outcomes for COVID-19 can be predicted using AI-driven models.

Autonomous control model for emergency operation of small modular reactor

Advancements in computer performance and Artificial Intelligence, especially Deep Neural Networks and reinforcement learning, have caught the attention of many researchers. Specifically, the potential for achieving autonomy (a higher cognitive concept than legacy automation systems) has garnered significant interest. Thus, many autonomous systems like self-driving cars, are in the research/ development phase. Despite its numerous advantages, the adoption of autonomous control for Nuclear Power Plants is deficient mainly due to stringent safety concerns. Especially, autonomous control for Small Modular Reactors is a desirous capability to accrue its full potential. This study proposes an AI-based model for autonomous control of IP-200 NPP, a Small Modular Reactor (SMR) under development. The model adopts a modular architecture for performing different functions of monitoring/ diagnosis, strategy formulation and assessment. Despite following knowledge-based rules, the network decisions are optimized through supervised learning and fine-tuned through reinforcement learning. The results demonstrate that the proposed model can handle plant emergencies while ensuring plant safety at optimal performance.