Automated Artificial Intelligence Research Articles

Prediction of coronary artery bypass graft outcomes using a single surgical note: An artificial intelligence-based prediction model study.

Healthcare providers currently calculate risk of the composite outcome of morbidity or mortality associated with a coronary artery bypass grafting (CABG) surgery through manual input of variables into a logistic regression-based risk calculator. This study indicates that automated artificial intelligence (AI)-based techniques can instead calculate risk. Specifically, we present novel numerical embedding techniques that enable NLP (natural language processing) models to achieve higher performance than the risk calculator using a single preoperative surgical note. The most recent preoperative surgical consult notes of 1,738 patients who received an isolated CABG from July 1, 2014 to November 1, 2022 at a single institution were analyzed. The primary outcome was the Society of Thoracic Surgeons defined composite outcome of morbidity or mortality (MM). We tested three numerical-embedding techniques on the widely used TextCNN classification model: 1a) Basic embedding, treat numbers as word tokens; 1b) Basic embedding with a dataloader that Replaces out-of-context (ROOC) numbers with a tag, where context is defined as within a number of tokens of specified keywords; 2) ScaleNum, an embedding technique that scales in-context numbers via a learned sigmoid-linear-log function; and 3) AttnToNum, a ScaleNum-derivative that updates the ScaleNum embeddings via multi-headed attention applied to local context. Predictive performance was measured via area under the receiver operating characteristic curve (AUC) on holdout sets from 10 random-split experiments. For eXplainable-AI (X-AI), we calculate SHapley Additive exPlanation (SHAP) values at an ngram resolution (SHAP-N). While the analyses focus on TextCNN, we execute an analogous performance pipeline with a long short-term memory (LSTM) model to test if the numerical embedding advantage is robust to model architecture. A total of 567 (32.6%) patients had MM following CABG. The embedding performances are as follows with the TextCNN architecture: 1a) Basic, mean AUC 0.788 [95% CI (confidence interval): 0.768-0.809]; 1b) ROOC, 0.801 [CI: 0.788-0.815]; 2) ScaleNum, 0.808 [CI: 0.785-0.821]; and 3) AttnToNum, 0.821 [CI: 0.806-0.834]. The LSTM architecture produced a similar trend. Permutation tests indicate that AttnToNum outperforms the other embedding techniques, though not statistically significant verse ScaleNum (p-value of .07). SHAP-N analyses indicate that the model learns to associate low blood urine nitrate (BUN) and creatinine values with survival. A correlation analysis of the attention-updated numerical embeddings indicates that AttnToNum learns to incorporate both number magnitude and local context to derive semantic similarities. This research presents both quantitative and clinical novel contributions. Quantitatively, we contribute two new embedding techniques: AttnToNum and ScaleNum. Both can embed strictly positive and bounded numerical values, and both surpass basic embeddings in predictive performance. The results suggest AttnToNum outperforms ScaleNum. With regards to clinical research, we show that AI methods can predict outcomes after CABG using a single preoperative note at a performance that matches or surpasses the current risk calculator. These findings reveal the potential role of NLP in automated registry reporting and quality improvement.

Improving sensitivity through data augmentation with synthetic lymph node metastases for AI-based analysis of PSMA PET-CT images.

We developed a fully automated artificial intelligence(AI)AI-based-based method for detecting suspected lymph node metastases in prostate-specific membrane antigen (PSMA)(PSMA) positron emission tomography-computed tomography (PET-CT)(PET-CT) images of prostate cancer patients by using data augmentation that adds synthetic lymph node metastases to the images to expand the training set. Synthetic data were derived from original training images to which new synthetic lymph node metastases were added. Thus, the original training set from a previous study (n = 420) was expanded by one synthetic image for every original image (n = 840), which was used to train an AI model. The performance of the AI model was compared to that of nuclear medicine physicians and a previously developed AI model. The human readers were alternately used as a reference and compared to either another reading or AI model. The new AI model had an average sensitivity of 84% for detecting lymph node metastases compared with 78% for human readings. Our previously developed AI method without synthetic data had an average sensitivity of 79%. The number of false positive lesions were slightly higher for the new AI model (average 3.3 instances per patient) compared to human readings and the previous AI model (average 2.8 instances per patient), while the number of false negative lesions was lower. Creating synthetic lymph node metastases, as a form of data augmentation, on [18F]PSMA-1007F]PSMA-1007 PETPET-CT-CT images improved the sensitivity of an AI model for detecting suspected lymph node metastases. However, the number of false positive lesions increased somewhat.

The Development and Potential Applications of an Automated Method for Detecting and Classifying Continuous Glucose Monitoring Patterns.

Continuous glucose monitoring (CGM) is emerging as a transformative tool for helping people with diabetes self-manage their glucose and supporting clinicians in effective treatment. Unfortunately, many CGM users, and clinicians, find interpreting the large volume of CGM data to be overwhelming and complex. To address this challenge, an efficient, intelligent method for detecting and classifying discernable patterns in CGM data was desired. We developed an automated artificial intelligence (AI)-driven method to detect and classify different discernable CGM patterns which called "CGM events." We trained different models using 60 days of CGM data from 27 individuals with diabetes from a publicly available data set and then evaluated model performance using separate test data from the same group. Each event is classified according to clinical significance based on three parameters: (1) glucose category at or near the beginning of the CGM event; (2) a calculated severity score that encompasses both signal shape and temporal characteristics (e.g., how high the CGM curve goes (measured in mg/dL) and how long it stays above target (as established by published consensus guidelines); and (3) the glucose category at or near the end of the event. The system accurately detected and classified events from actual CGM data. This was also validated with expert diabetes clinicians. Advanced pattern recognition methods can be used to detect and classify CGM events of interest and may be used to provide actionable insights and self-management support to CGM users and decision support to the clinicians caring for them.

AI-generated CT body composition biomarkers associated with increased mortality risk in socioeconomically disadvantaged individuals.

To evaluate the relationship between socioeconomic disadvantage using national area deprivation index (ADI) and CT-based body composition measures derived from fully automated artificial intelligence (AI) tools to identify body composition measures associated with increased risk for all-cause mortality and adverse cardiovascular events. Fully automated AI body composition tools quantifying abdominal aortic calcium, abdominal fat (visceral [VAT], visceral-to-subcutaneous ratio [VSR]), and muscle attenuation (muscle HU) were applied to non-contrast CT examinations in adults undergoing screening CT colonography (CTC). Patients were partitioned into 5 socioeconomic groups based on the national ADI rank at the census block group level. Pearson correlation analysis was performed to determine the association between national ADI and body composition measures. One-way analysis of variance was used to compare means across groups. Odds ratios (ORs) were generated using high-risk, high specificity (90% specificity) body composition thresholds with the most disadvantaged groups being compared to the least disadvantaged group (ADI < 20). 7785 asymptomatic adults (mean age, 57years; 4361:3424 F:M) underwent screening CTC from April 2004-December 2016. ADI rank data were available in 7644 patients. Median ADI was 31 (IQR 22-43). Aortic calcium, VAT, and VSR had positive correlation with ADI and muscle attenuation had a negative correlation with ADI (all p < .001). Compared with the least disadvantaged group, mean differences for the most disadvantaged group (ADI > 80) were: Aortic calcium (Agatston) = 567, VAT = 27 cm2, VSR = 0.1, and muscle HU = -6 HU (all p < .05). Compared with the least disadvantaged group, the most disadvantaged group had significantly higher odds of having high-risk body composition measures: Aortic calcium OR = 3.8, VAT OR = 2.5, VSR OR = 2.0, and muscle HU OR = 3.1(all p < .001). Fully automated CT body composition tools show that socioeconomic disadvantage is associated with high-risk body composition measures and can be used to identify individuals at increased risk for all-cause mortality and adverse cardiovascular events.

Machine learning model of facial expression outperforms models using analgesia nociception index and vital signs to predict postoperative pain intensity: a pilot study.

Few studies have evaluated the use of automated artificial intelligence (AI)-based pain recognition in postoperative settings or the correlation with pain intensity. In this study, various machine learning (ML)-based models using facial expressions, the analgesia nociception index (ANI), and vital signs were developed to predict postoperative pain intensity, and their performances for predicting severe postoperative pain were compared. In total, 155 facial expressions from patients who underwent gastrectomy were recorded postoperatively; one blinded anesthesiologist simultaneously recorded the ANI score, vital signs, and patient self-assessed pain intensity based on the 11-point numerical rating scale (NRS). The ML models' area under the receiver operating characteristic curves (AUROCs) were calculated and compared using DeLong's test. ML models were constructed using facial expressions, ANI, vital signs, and different combinations of the three datasets. The ML model constructed using facial expressions best predicted an NRS ≥ 7 (AUROC 0.93) followed by the ML model combining facial expressions and vital signs (AUROC 0.84) in the test-set. ML models constructed using combined physiological signals (vital signs, ANI) performed better than models based on individual parameters for predicting NRS ≥ 7, although the AUROCs were inferior to those of the ML model based on facial expressions (all P < 0.050). Among these parameters, absolute and relative ANI had the worst AUROCs (0.69 and 0.68, respectively) for predicting NRS ≥ 7. The ML model constructed using facial expressions best predicted severe postoperative pain (NRS ≥ 7) and outperformed models constructed from physiological signals.

Machine learning based hybrid anomaly detection technique for automatic diagnosis of cardiovascular diseases using cardiac sympathetic nerve activity and electrocardiogram.

Coronary artery diseases (CADs) are the leading cause of death worldwide and early diagnosis is crucial for timely treatment. To address this, our study presents a novel automated Artificial Intelligence (AI)-based Hybrid Anomaly Detection (AIHAD) technique that combines various signal processing, feature extraction, supervised, and unsupervised machine learning methods. By jointly and simultaneously analyzing 12-lead cardiac sympathetic nerve activity (CSNA) and electrocardiogram (ECG) data, the automated AIHAD technique performs fast, early, and accurate diagnosis of CADs. In order to develop and evaluate the proposed automated AIHAD technique, we utilized the fully labeled STAFF III and PTBD databases, which contain the 12-lead wideband raw recordings non-invasively acquired from 260 subjects. Using these wideband raw recordings, we developed a signal processing technique that simultaneously detects the 12-lead CSNA and ECG signals of all subjects. Using the pre-processed 12-lead CSNA and ECG signals, we developed a time-domain feature extraction technique that extracts the statistical CSNA and ECG features critical for the reliable diagnosis of CADs. Using the extracted discriminative features, we developed a supervised classification technique based on Artificial Neural Networks (ANNs) that simultaneously detects anomalies in the 12-lead CSNA and ECG data. Furthermore, we developed an unsupervised clustering technique based on Gaussian mixture models (GMMs) and Neyman-Pearson criterion, which robustly detects outliers corresponding to CADs. Using the automated AIHAD technique, we have, for the first time, demonstrated a significant association between the increase in CSNA signals and anomalies in ECG signals during CADs. The AIHAD technique achieved highly reliable detection of CADs with a sensitivity of 98.48 %, specificity of 97.73 %, accuracy of 98.11 %, positive predictive value of 97.74 %, negative predictive value of 98.47 %, and F1-score of 98.11 %. Hence, the automated AIHAD technique demonstrates superior performance compared to the gold standard diagnostic test ECG in the diagnosis of CADs. Additionally, it outperforms other techniques developed in this study that separately utilize either only CSNA data or only ECG data. Therefore, it significantly increases the detection performance of CADs by taking advantage of the diversity in different data types and leveraging their strengths. Furthermore, its performance is comparatively better than that of most previously proposed machine and deep learning methods that exclusively used ECG data to diagnose or classify CADs. Additionally, it has a very low implementation time, which is highly desirable for real-time detection of CADs. The proposed automated AIHAD technique may serve as an efficient decision-support system to increase physicians' success in fast, early, and accurate diagnosis of CADs. It may be highly beneficial and valuable, particularly for asymptomatic patients, for whom the diagnostic information provided by ECG alone is not sufficient to reliably diagnose the disease. Hence, it may significantly improve patient outcomes by enabling timely treatments and considerably reducing the mortality of cardiovascular diseases (CVDs).

Detection of Moderate Hepatic Steatosis on Portal Venous Phase Contrast-Enhanced CT: Evaluation Using an Automated Artificial Intelligence Tool.

BACKGROUND. Precontrast CT is an established means of evaluating for hepatic steatosis; postcontrast CT has historically been limited for this purpose. OBJECTIVE. The purpose of this study was to evaluate the diagnostic performance of portal venous phase postcontrast CT in detecting at least moderate hepatic steatosis using liver and spleen attenuation measurements determined by an automated artificial intelligence (AI) tool. METHODS. This retrospective study included 2917 patients (1381 men, 1536 women; mean age, 56.8 years) who underwent a CT examination that included at least two series through the liver. Examinations were obtained from an AI vendor's data lake of data from 24 centers in one U.S. health care network and 29 centers in one Israeli health care network. An automated deep learning tool extracted liver and spleen attenuation measurements. The reference for at least moderate steatosis was precontrast liver attenuation of less than 40 HU (i.e., estimated liver fat > 15%). A radiologist manually reviewed examinations with outlier AI results to confirm portal venous timing and identify issues impacting attenuation measurements. RESULTS. After outlier review, analysis included 2777 patients with portal venous phase images. Prevalence of at least moderate steatosis was 13.9% (387/2777). Patients without and with at least moderate steatosis, respectively, had mean postcontrast liver attenuation of 104.5 ± 18.1 (SD) HU and 67.1 ± 18.6 HU (p < .001); a mean difference in postcontrast attenuation between the liver and the spleen (hereafter, postcontrast liver-spleen attenuation difference) of -7.6 ± 16.4 (SD) HU and -31.8 ± 20.3 HU (p < .001); and mean liver enhancement of 49.3 ± 15.9 (SD) HU versus 38.6 ± 13.6 HU (p < .001). Diagnostic performance for the detection of at least moderate steatosis was higher for postcontrast liver attenuation (AUC = 0.938) than for the postcontrast liver-spleen attenuation difference (AUC = 0.832) (p < .001). For detection of at least moderate steatosis, postcontrast liver attenuation had sensitivity and specificity of 77.8% and 93.2%, respectively, at less than 80 HU and 90.5% and 78.4%, respectively, at less than 90 HU; the postcontrast liver-spleen attenuation difference had sensitivity and specificity of 71.4% and 79.3%, respectively, at less than -20 HU and 87.0% and 62.1%, respectively, at less than -10 HU. CONCLUSION. Postcontrast liver attenuation outperformed the postcontrast liver-spleen attenuation difference for detecting at least moderate steatosis in a heterogeneous patient sample, as evaluated using an automated AI tool. Splenic attenuation likely is not needed to assess for at least moderate steatosis on postcontrast images. CLINICAL IMPACT. The technique could promote early detection of clinically significant nonalcoholic fatty liver disease through individualized or large-scale opportunistic evaluation.

Evaluation of an artificial intelligence-based system for echocardiographic estimation of right atrial pressure

Current noninvasive estimation of right atrial pressure (RAP) by inferior vena cava (IVC) measurement during echocardiography may have significant inter-rater variability due to different levels of observers’ experience. Therefore, there is a need to develop new approaches to decrease the variability of IVC analysis and RAP estimation. This study aims to develop a fully automated artificial intelligence (AI)-based system for automated IVC analysis and RAP estimation. We presented a multi-stage AI system to identify the IVC view, select good quality images, delineate the IVC region and quantify its thickness, enabling temporal tracking of its diameter and collapsibility changes. The automated system was trained and tested on expert manual IVC and RAP reference measurements obtained from 255 patients during routine clinical workflow. The performance was evaluated using Pearson correlation and Bland-Altman analysis for IVC values, as well as macro accuracy and chi-square test for RAP values. Our results show an excellent agreement (r=0.96) between automatically computed versus manually measured IVC values, and Bland-Altman analysis showed a small bias of -\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$-$$\\end{document}0.33 mm. Further, there is an excellent agreement ((p<0.01\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$(p<0.01$$\\end{document}) between automatically estimated versus manually derived RAP values with a macro accuracy of 0.85. The proposed AI-based system accurately quantified IVC diameter, collapsibility index, both are used for RAP estimation. This automated system could serve as a paradigm to perform IVC analysis in routine echocardiography and support various cardiac diagnostic applications.

Assessing the role of an artificial intelligence assessment tool for thoracic aorta diameter on routine chest CT.

To assess the diagnostic accuracy and clinical impact of automated artificial intelligence (AI) measurement of thoracic aorta diameter on routine chest CT. A single-centre retrospective study involving three cohorts. 210 consecutive ECG-gated CT aorta scans (mean age 75 ± 13) underwent automated analysis (AI-Rad Companion Chest CT, Siemens) and were compared to a reference standard of specialist cardiothoracic radiologists for accuracy measuring aortic diameter. A repeated measures analysis tested reporting consistency in a second cohort (29 patients, mean age 61 ± 17) of immediate sequential pre-contrast and contrast CT aorta acquisitions. Potential clinical impact was assessed in a third cohort of 197 routine CT chests (mean age 66 ± 15) to document potential clinical impact. AI analysis produced a full report in 387/436 (89%) and a partial report in 421/436 (97%). Manual vs AI agreement was good to excellent (ICC 0.76-0.92). Repeated measures analysis of expert and AI reports for the ascending aorta were moderate to good (ICC 0.57-0.88). AI diagnostic performance crossed the the threshold for maximally accepted limits of agreement (>5 mm) at the aortic root on ECG-gated CTs. AI newly identified aortic dilatation in 27% of patients on routine thoracic imaging with a specificity of 99% and sensitivity of 77%. AI has good agreement with expert readers at the mid ascending aorta and has high specificity, but low sensitivity, at detecting dilated aortas on non-dedicated chest CTs. An AI tool may improve the detection of previously unknown thoracic aorta dilatation on chest CTs versus current routine reporting.

Deep learning-based prediction model for diagnosing gastrointestinal diseases using endoscopy images

BackgroundGastrointestinal (GI) infections are quite common today around the world. Colonoscopy or wireless capsule endoscopy (WCE) are noninvasive methods for examining the whole GI tract for abnormalities. Nevertheless, it requires a great deal of time and effort for doctors to visualize a large number of images, and diagnosis is prone to human error. As a result, developing automated artificial intelligence (AI) based GI disease diagnosis methods is a crucial and emerging research area. AI-based prediction models may lead to improvements in the early diagnosis of gastrointestinal disorders, assessing severity, and healthcare systems for the benefit of patients as well as clinicians. The focus of this research is on the early diagnosis of gastrointestinal diseases using a convolution neural network (CNN) to enhance diagnosis accuracy. MethodsVarious CNN models (baseline model and using transfer learning (VGG16, InceptionV3, and ResNet50)) were trained on a benchmark image dataset, KVASIR, containing images from inside the GI tract using n-fold cross-validation. The dataset comprises images of three disease states—polyps, ulcerative colitis, and esophagitis—as well as images of the healthy colon. Data augmentation strategies together with statistical measures were used to improve and evaluate the model's performance. Additionally, the test set comprising 1200 images was used to evaluate the model's accuracy and robustness. ResultsThe CNN model using the weights of the ResNet50 pre-trained model achieved the highest average accuracy of approximately 99.80% on the training set (100% precision and approximately 99% recall) and accuracies of 99.50% and 99.16% on the validation and additional test set, respectively, while diagnosing GI diseases. When compared to other existing systems, the proposed ResNet50 model outperforms them all. ConclusionThe findings of this study indicate that AI-based prediction models using CNNs, specifically ResNet50, can improve diagnostic accuracy for detecting gastrointestinal polyps, ulcerative colitis, and esophagitis. The prediction model is available at https://github.com/anjus02/GI-disease-classification.git.

O-239 A model based on artificial intelligence for the non-invasive prediction of embryo aneuploidy

Abstract Study question Can the automated artificial intelligence (AI) model iDAScore predict embryo aneuploidy? Summary answer The AUC for aneuploidy prediction of iDAScore was AUC 0.640, with optimal cut-off score 7.75, sensitivity 61.0% and specificity 60.9%. What is known already Embryo morphokinetics and image analysis using AI have been proposed as alternative approaches for non-invasive embryo aneuploidy prediction. The automated AI model iDAScore, as well as the morphokinetic embryo selection model KIDScoreD5 have been shown to have correlation with embryo ploidy status. The aim was to study the performance of aneuploidy prediction using the iDAScore, KIDScoreD5 and morphology grading (Gardner criteria). Study design, size, duration Retrospective study including 382 blastocysts of known ploidy from 92 patients undergoing PGT-A in a single centre during January-December 2022. For statistical analysis, continuous and categorical data were compared using ANOVA and Fisher’s exact test/Chi-square, respectively. ROC analysis was used to estimate the discrimination performance of variables to predict aneuploidy. Stepwise logistic regression was used to identify confounding variables and construct a model for aneuploidy prediction. Participants/materials, setting, methods Embryos were cultured in Embryoscope, underwent laser assisted hatching on Day3, were biopsied on Day5-Day6 and were vitrified. Whole-genome amplification and next-generation sequencing were used for PGT-A. iDAScore values were calculated using v1.2.0. KIDScoreD5 v.3 was calculated after annotation of the required morphokinetic parameters in Embryoviewer. For Gardner criteria, inner cell mass (ICM) and trophectoderm (TE) were classified into grades A to C. The expansion in all biopsied blastocysts was &amp;gt; =grade 4. Main results and the role of chance Out of 382 embryos analysed, 127 (33.2%) were euploid and 255 (66.8%) were aneuploid. Euploid embryos were associated with significantly higher iDAScore (7.69, 95% CI 7.44-7.95 vs 6.86, 95% CI 6.66-7.09, p &amp;lt; 0.001), higher KIDScoreD5 (6.41, 95% CI 6.07-6.75 vs 5.74, 95% CI 5.51-5.97, p = 0.001) and lower maternal age (37.10 years, 95% 36.39-37.82 vs 39.83 years, 95% CI 39.36-40.30, p &amp;lt; 0.001) compared to aneuploid embryos. Significantly more embryos with grade A ICM (A:39.4%, B:26.0%, C:20.0%, p = 0.013) and grade A TE (A: 45.9%, B:29.6%, C:23.6%, p = 0.002) were euploid compared with B and C grades, respectively. Blastocysts biopsied on Day 5 and Day 6 had similar proportions of euploidy (34.9% vs 27.6%, p = 0.125). The AUCs for aneuploidy prediction were: iDAScore: 0.640 (95% CI 0.582-0.699), KIDScoreD5: 0.604 (95% CI 0.542-0.665), ICM grade: 0.581 (95% CI 0.521-0.640), TE grade: 0.596 (0.536-0.657), maternal age: 0.694 (0.640-0.748). The optimal cut-off (Youdex index) for aneuploidy prediction of iDAScore was 7.75 (sensitivity 61.0%, specificity 60.9%). Following stepwise multivariate logistic regression, the variables that entered the model were iDAScore, maternal age and TE grade. The model iDAScore+age+TE grade increased the ability of aneuploidy prediction (AUC 0.747, 95% CI 0.696-0.796). KIDScoreD5, ICM grade and day of biopsy failed to enter the model. Limitations, reasons for caution This is a retrospective study and therefore the presence of bias cannot be excluded. iDAScore and KIDScore are models originally developed for the prediction of implantation and not for the diagnosis of embryo aneuploidy. Wider implications of the findings iDAScore appears to associate with embryo aneuploidy. Combination of iDAScore+age+TE increased the AUC for aneuploidy prediction. iDAScore can used as a decision-support tool for prioritising embryos for transfer, cryopreservation or biopsy. The analysis of more embryos is needed to confirm the present findings. Trial registration number not applicable

Artificial intelligence-based diagnosis of breast cancer by mammography microcalcification

Mammography is the mainstream imaging modality used for breast cancer screening. Identification of microcalcifications associated with malignancy may result in early diagnosis of breast cancer and aid in reducing the morbidity and mortality associated with the disease. Computer-aided diagnosis (CAD) is a promising technique due to its efficiency and accuracy. Here, we demonstrated that an automated deep-learning pipeline for microcalcification detection and classification on mammography can facilitate early diagnosis of breast cancer. This technique can not only provide the classification results of mammography, but also annotate specific calcification regions. A large mammography dataset was collected, including 4,810 mammograms with 6,663 microcalcification lesions based on biopsy results, of which 3,301 were malignant and 3,362 were benign. The system was developed and tested using images from multiple centers. The overall classification accuracy values for discriminating between benign and malignant breasts were 0.8124 for the training set and 0.7237 for the test set. The sensitivity values of malignant breast cancer prediction were 0.8891 for the training set and 0.7778 for the test set. In addition, we collected information regarding pathological sub-type (pathotype) and estrogen receptor (ER) status, and we subsequently explored the effectiveness of deep learning-based pathotype and ER classification. Automated artificial intelligence (AI) systems may assist clinicians in making judgments and improve their efficiency in breast cancer screening, diagnosis, and treatment.

Incremental prognostic value of fully-automatic LVEF by stress CMR using machine learning

Abstract Funding Acknowledgements Type of funding sources: None. Background Although some recent reports showed that fully automated artificial intelligence (AI)-based left ventricular ejection fraction (LVEF) measured at stress has good performance, its prognostic value during a stress CMR exam to predict outcomes is not well established. Aim To determine in patients undergoing stress CMR whether fully automated AI-based LVEF (LVEF-AI) measured at stress can provide incremental prognostic value to predict death. Methods Between 2016 and 2018, we conducted a longitudinal study including all consecutive patients referred for vasodilator stress CMR. LVEF-AI was assessed using AI-algorithm combines multiple deep learning networks for LV segmentation. The primary outcome was all-cause death defined by the French National Registry of Death. Cox regression was used to evaluate the association of stress LVEF-AI with death after adjustment for traditional risk factors. Results In 9,712 patients (66±15 years, 67% men), there was an excellent correlation between stress LVEF-AI measurement and LVEF measured by expert (LVEF-expert) (r=0.94, p&amp;lt;0.001). Using Bland–Altman analysis, we found that the difference between the mean LVEF-expert and the LVEF-AI group was −0.1% (−0.066–0.067), that was not statistically significant (p = 0.46). Stress LVEF-AI was associated with death (median [IQR] follow-up 4.5 [3.7–5.2] years) before and after adjustment for risk factors (adjusted hazard ratio [HR], 0.84 [95% CI, 0.82–0.87] per 5% increment, p&amp;lt;0.001). Stress LVEF-AI had similar significant association with death occurrence compared with LVEF-expert. After adjustment, a lower stress LVEF-AI showed the greatest improvement in model discrimination and reclassification over and above traditional risk factors and stress CMR findings (C-statistic improvement: 0.11; NRI=0.250; IDI=0.049, all p&amp;lt;0.001; LR-test p&amp;lt;0.001), with a superior additional prognostic value than the LVEF-AI measured at rest. Conclusion AI-based fully automated LVEF measured at stress is independently associated with the occurrence of death in patients undergoing stress CMR, with an additional prognostic value above traditional risk factors, inducible ischemia and LGE.

CT image-based biomarkers acquired by AI-based algorithms for the opportunistic prediction of falls.

Evaluate whether biomarkers measured by automated artificial intelligence (AI)-based algorithms are suggestive of future fall risk. In this retrospective age- and sex-matched case-control study, 9029 total patients underwent initial abdominal CT for a variety of indications over a 20-year interval at one institution. 3535 case patients (mean age at initial CT, 66.5 ± 9.6 years; 63.4% female) who went on to fall (mean interval to fall, 6.5 years) and 5494 controls (mean age at initial CT, 66.7 ± 9.8 years; 63.4% females; mean follow-up interval, 6.6 years) were included. Falls were identified by electronic health record review. Validated and fully automated quantitative CT algorithms for skeletal muscle, adipose tissue, and trabecular bone attenuation at the level of L1 were applied to all scans. Uni- and multivariate assessment included hazard ratios (HRs) and area under the receiver operating characteristic (AUROC) curve. Fall HRs (with 95% CI) for low muscle Hounsfield unit, high total adipose area, and low bone Hounsfield unit were 1.82 (1.65-2.00), 1.31 (1.19-1.44) and 1.91 (1.74-2.11), respectively, and the 10-year AUROC values for predicting falls were 0.619, 0.556, and 0.639, respectively. Combining all these CT biomarkers further improved the predictive value, including 10-year AUROC of 0.657. Automated abdominal CT-based opportunistic measures of muscle, fat, and bone offer a novel approach to risk stratification for future falls, potentially by identifying patients with osteosarcopenic obesity. There are few well-established clinical tools to predict falls. We use novel AI-based body composition algorithms to leverage incidental CT data to help determine a patient's future fall risk.

Fully automated calcium scoring predicts all-cause mortality at 12 years in the MILD lung cancer screening trial.

Coronary artery calcium (CAC) is a known risk factor for cardiovascular (CV) events and mortality but is not yet routinely evaluated in low-dose computed tomography (LDCT)-based lung cancer screening (LCS). The present analysis explored the capacity of a fully automated CAC scoring to predict 12-year mortality in the Multicentric Italian Lung Detection (MILD) LCS trial. The study included 2239 volunteers of the MILD trial who underwent a baseline LDCT from September 2005 to January 2011, with a median follow-up of 190 months. The CAC score was measured by a commercially available fully automated artificial intelligence (AI) software and stratified into five strata: 0, 1-10, 11-100, 101-400, and > 400. Twelve-year all-cause mortality was 8.5% (191/2239) overall, 3.2% with CAC = 0, 4.9% with CAC = 1-10, 8.0% with CAC = 11-100, 11.5% with CAC = 101-400, and 17% with CAC > 400. In Cox proportional hazards regression analysis, CAC > 400 was associated with a higher 12-year all-cause mortality both in a univariate model (hazard ratio, HR, 5.75 [95% confidence interval, CI, 2.08-15.92] compared to CAC = 0) and after adjustment for baseline confounders (HR, 3.80 [95%CI, 1.35-10.74] compared to CAC = 0). All-cause mortality significantly increased with increasing CAC (7% in CAC ≤ 400 vs. 17% in CAC > 400, Log-Rank p-value <0.001). Non-cancer at 12 years mortality was 3% (67/2239) overall, 0.8% with CAC = 0, 1.0% with CAC = 1-10, 2.9% with CAC = 11-100, 3.6% with CAC = 101-400, and 8.2% with CAC > 400 (Grey's test p < 0.001). In Fine and Gray's competing risk model, CAC > 400 predicted 12-year non-cancer mortality in a univariate model (sub-distribution hazard ratio, SHR, 10.62 [95% confidence interval, CI, 1.43-78.98] compared to CAC = 0), but the association was no longer significant after adjustment for baseline confounders. In conclusion, fully automated CAC scoring was effective in predicting all-cause mortality at 12 years in a LCS setting.

A Novel Approach for Brain Tumor Classification Using an Ensemble of Deep and Hand-Crafted Features.

One of the most severe types of cancer caused by the uncontrollable proliferation of brain cells inside the skull is brain tumors. Hence, a fast and accurate tumor detection method is critical for the patient's health. Many automated artificial intelligence (AI) methods have recently been developed to diagnose tumors. These approaches, however, result in poor performance; hence, there is a need for an efficient technique to perform precise diagnoses. This paper suggests a novel approach for brain tumor detection via an ensemble of deep and hand-crafted feature vectors (FV). The novel FV is an ensemble of hand-crafted features based on the GLCM (gray level co-occurrence matrix) and in-depth features based on VGG16. The novel FV contains robust features compared to independent vectors, which improve the suggested method's discriminating capabilities. The proposed FV is then classified using SVM or support vector machines and the k-nearest neighbor classifier (KNN). The framework achieved the highest accuracy of 99% on the ensemble FV. The results indicate the reliability and efficacy of the proposed methodology; hence, radiologists can use it to detect brain tumors through MRI (magnetic resonance imaging). The results show the robustness of the proposed method and can be deployed in the real environment to detect brain tumors from MRI images accurately. In addition, the performance of our model was validated via cross-tabulated data.

PD13-13 A FULLY AUTOMATED ARTIFICIAL INTELLIGENCE SUPPORT SYSTEM FOR URINE CYTOLOGY: MULTI-CENTER EXTERNAL VALIDATION STUDY

You have accessJournal of UrologyCME1 Apr 2023PD13-13 A FULLY AUTOMATED ARTIFICIAL INTELLIGENCE SUPPORT SYSTEM FOR URINE CYTOLOGY: MULTI-CENTER EXTERNAL VALIDATION STUDY Masatomo Kaneko, Keisuke Tsuji, Yuki Harada, Atsuko Fujihara, Kengo Ueno, Masaya Nakanishi, Eiichi Konishi, Tetsuro Takamatsu, Satoshi Teramukai, Toshiko Ito-Ihara, Andre Abreu, and Osamu Ukimura Masatomo KanekoMasatomo Kaneko More articles by this author , Keisuke TsujiKeisuke Tsuji More articles by this author , Yuki HaradaYuki Harada More articles by this author , Atsuko FujiharaAtsuko Fujihara More articles by this author , Kengo UenoKengo Ueno More articles by this author , Masaya NakanishiMasaya Nakanishi More articles by this author , Eiichi KonishiEiichi Konishi More articles by this author , Tetsuro TakamatsuTetsuro Takamatsu More articles by this author , Satoshi TeramukaiSatoshi Teramukai More articles by this author , Toshiko Ito-IharaToshiko Ito-Ihara More articles by this author , Andre AbreuAndre Abreu More articles by this author , and Osamu UkimuraOsamu Ukimura More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003260.13AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: To validate a fully automated artificial intelligence (AI) support system for urine cytology to detect histological high-grade urothelial carcinoma (HGUC). METHODS: A total of 660 urine cytology slides were collected from the urine of consecutive patients just before bladder biopsy or transurethral resection of the bladder at 3 institutions (center 1, Kyoto Prefectural University of Medicine Hospital; center 2, Japanese Red Cross Kyoto Second Hospital; center 3, North Medical Center Kyoto Prefectural University of Medicine) between 2016 to 2022 (IRB ERB-C 1339-5, 1673-1, and S2020-60). Board-certified cytotechnologists and pathologist independently labeled each cell and classified each slide according to the Paris system (TPS). After excluding an unsatisfactory slide, cytology slides were digitized for image analysis. AI was developed with a deep learning method using 181 slides obtained in center 1. Urinary cells on a slide were automatically detected by AI, each cell was classified into benign vs malignant, the results of cell-level classifications were integrated, and AI classified each slide into negative (benign or atypical urothelial cells) vs positive (suspicious HGUC or HGUC). The model showing the highest accuracy in the training dataset was selected, and the slide classification performance was tested with 315 slides obtained from center 1 for internal validation and 163 slides obtained from centers 2 and 3 for external validation. The diagnostic performance to detect histological HGUC assessed with TPS and slide classification with AI were compared. The receiver operating characteristic (ROC) analysis was performed for the binary classification. Statistical significance defined as p<0.05. RESULTS: For slide classification, the area under the ROC (AUC) of AI was 0.93 in internal validation and 0.84 in external validation. The accuracy was 85% in both internal and external datasets. For diagnostic performance to detect histological HGUC, AUC was 0.78 in internal validation and 0.73 in external validation. The accuracy/sensitivity/specificity with TPS was 68%/46%/89% in internal dataset and 66%/35%/91% in external dataset. When the sensitivity was matched with TPS, the accuracy of AI was 68% (p=1.0) and 66% (p=1.0) and specificity was 89% (p=0.6) and 91% (p=0.7) for internal and external validation, respectively. CONCLUSIONS: A fully automated AI system accurately classifies the digitized urine cytology slides and detects histological HGUC with comparative performance to TPS in both internal and external validation. Source of Funding: None © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e410 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Masatomo Kaneko More articles by this author Keisuke Tsuji More articles by this author Yuki Harada More articles by this author Atsuko Fujihara More articles by this author Kengo Ueno More articles by this author Masaya Nakanishi More articles by this author Eiichi Konishi More articles by this author Tetsuro Takamatsu More articles by this author Satoshi Teramukai More articles by this author Toshiko Ito-Ihara More articles by this author Andre Abreu More articles by this author Osamu Ukimura More articles by this author Expand All Advertisement PDF downloadLoading ...

MP30-02 A NOVEL PREDICTION MODEL FOR POSSIBILITY OF SPERM RETRIEVAL IN MICRO TESE FOR NON-OBSTRUCTIVE AZOOSPERMIA PATIENTS USING AUTOMATED ARTIFICIAL INTELLIGENCE

You have accessJournal of UrologyCME1 Apr 2023MP30-02 A NOVEL PREDICTION MODEL FOR POSSIBILITY OF SPERM RETRIEVAL IN MICRO TESE FOR NON-OBSTRUCTIVE AZOOSPERMIA PATIENTS USING AUTOMATED ARTIFICIAL INTELLIGENCE Hideyuki Kobayashi, Masato Uetani, Koichi Nakajima, and Koichi Nagao Hideyuki KobayashiHideyuki Kobayashi More articles by this author , Masato UetaniMasato Uetani More articles by this author , Koichi NakajimaKoichi Nakajima More articles by this author , and Koichi NagaoKoichi Nagao More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003258.02AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Azoospermia is a severe problem that prevents couples from having their own children through natural pregnancy. In non-obstructive azoospermia (NOA), a particularly severe type of male infertility, microdissection testicular sperm extraction (micro TESE) is required to collect sperm and, at 40-60%, the sperm retrieval success rate is not very high.Previous studies identified no single clinical finding or investigation that could accurately predict the outcome of sperm retrieval. In addition, although clinical and hormone parameters have been previously investigated as potential predictors of sperm retrieval, the evidence for them has been conflicting and a specific biochemical marker predicting the possibility of sperm retrieval has yet to be established. It would be very valuable to know the possibility of sperm retrieval for patients with NOA before performing micro TESE. Therefore, our objective was to make a model for predicting the possibility of sperm retrieval in patients with NOA before performing micro TESE, using machine learning. METHODS: We retrospectively obtained data from the medical records of 430 patients who underwent micro TESE from 2011 to 2020. Parameters extracted were age, height, body weight, body mass index, LH, FSH, FSH, PRL, total testosterone, E2, T/E2, sperm retrieval, G-band, AZF, medical history, Rt testis, and Lt testis. Prediction One, which does not require coding, was used to create an AI prediction model for sperm retrieval. Prediction One makes the best prediction model using an artificial neural network (ANN) with internal cross validation. RESULTS: The mean age of the 430 male patients who underwent micro TESE in this study was 36.78±7.58 years. Sperm was successfully retrieved in 151 (35.1%) patients.Statistically significant differences between successful and unsuccessful sperm retrieval were observed for age, LH, FSH, total testosterone, Rt testis, Lt testis and medical history. Medical history included none, Lt varicocele, cryptorchidism, inguinal hernia, torsion of spermatic cord, orchitis, cancer treatment, spinal injury, spina bifida, and anti-sperm antibody (+). The AUC for the AI model was 72.46%, which is acceptable. The values for Accuracy, Precision and Recall were 72.09%, 64.76% and 45.03%, respectively, when the F-value was 53.13%. CONCLUSIONS: We created an AI model for predicting sperm retrieval in patients with NOA before undergoing micro TESE. Source of Funding: none © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e391 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Hideyuki Kobayashi More articles by this author Masato Uetani More articles by this author Koichi Nakajima More articles by this author Koichi Nagao More articles by this author Expand All Advertisement PDF downloadLoading ...

An artificial intelligence platform provides an accurate interpretation of esophageal motility from Functional Lumen Imaging Probe Panometry studies.

Functional lumen imaging probe (FLIP) Panometry is performed at the time of sedated endoscopy and evaluates esophageal motility in response to distension. This study aimed to develop and test an automated artificial intelligence (AI) platform that could interpret FLIP Panometry studies. The study cohort included 678 consecutive patients and 35 asymptomatic controls that completed FLIP Panometry during endoscopy and high-resolution manometry (HRM). "True" study labels for model training and testing were assigned by experienced esophagologists per a hierarchical classification scheme. The supervised, deep learning, AI model generated FLIP Panometry heatmaps from raw FLIP data and based on convolutional neural networks assigned esophageal motility labels using a two-stage prediction model. Model performance was tested on a 15% held-out test set (n=103); the remainder of the studies were utilized for model training (n=610). "True" FLIP labels across the entire cohort included 190 (27%) "normal," 265 (37%) "not normal/not achalasia," and 258 (36%) "achalasia." On the test set, both the Normal/Not normal and the achalasia/not achalasia models achieved an accuracy of 89% (with 89%/88% recall, 90%/89% precision, respectively). Of 28 patients with achalasia (per HRM) in the test set, 0 were predicted as "normal" and 93% as "achalasia" by the AI model. An AI platform provided accurate interpretation of FLIP Panometry esophageal motility studies from a single center compared with the impression of experienced FLIP Panometry interpreters. This platform may provide useful clinical decision support for esophageal motility diagnosis from FLIP Panometry studies performed at the time of endoscopy.

A Deep Learning Algorithm for Automatic 3D Segmentation of Rotator Cuff Muscle and Fat from Clinical MRI Scans.

The authors aimed to develop and validate an automated artificial intelligence (AI) algorithm for three-dimensional (3D) segmentation of all four rotator cuff (RC) muscles to quantify intramuscular fat infiltration (FI) and individual muscle volume. The dataset included retrospectively collected RC MRI scans in 232 patients (63 with normal RCs, 169 with RC tears). A two-stage AI model was developed to segment all RC muscles and their FI in each stage. For comparison, single-stage and Otsu filtering models were created. Using the two-stage model, segmentation performance demonstrated high Dice scores (mean, 0.92 ± 0.14 [SD]), low volume errors (mean, 5.72% ± 9.23), and low FI errors (mean, 1.54% ± 2.79) when validated in 30 scans. There was a significant correlation between the 3D FI in the RC tear scans with a Goutallier grade (ρ = 0.53, P < .001) and FI found from a single two-dimensional (2D) section (all muscles, ρ > 0.70; P < .001). However, Bland-Altman analysis of the 3D compared with the 2D analyses of FI demonstrated a proportional bias (all muscles, P < .001). Compared with Goutallier classification or single-image quantification, the AI method allowed for more variability in images and led to objective separate quantifications of muscle volume and FI in all RC muscles. Keywords: Rotator Cuff, Artificial Intelligence, Segmentation, Fat Infiltration, Muscle Volume, MRI, Shoulder Supplemental material is available for this article. © RSNA, 2023.