Recent advancements in artificial intelligence have led to increased interest in predictive modeling across various domains, including medicine. Although numerous metrics have been established for binary classification, the growing adoption of multi-class systems necessitates robust evaluation measures. However, comprehensive simulation studies comparing the performance of existing multi-class metrics under diverse data conditions remain limited. In this study, we first provide a concise overview of commonly used accuracy metrics for multi-class classification. Then, we report a simulation study that systematically evaluates several diagnostic accuracy measures under a wide range of scenarios, including three- and five-class settings, balanced and imbalanced sample sizes, and different distributional assumptions for predictors. We assessed each metric’s performance in terms of bias and 95% confidence interval coverage. Under balanced conditions, most metrics demonstrated stable and unbiased performance, closely approximating the true values. However, under imbalanced conditions, greater bias was observed, with the M-index and polytomous discrimination index exhibiting comparatively more stable performance across various scenarios. The micro-averaged receiver operating characteristic curve area consistently showed higher bias under class imbalance. Finally, we applied these metrics to a glioma tumor grading task using external datasets. This study provides a systematic comparison of commonly used metrics and offers practical guidance for selecting appropriate measures in multi-class classification tasks.

Objective: This study aimed to assess the association between exposure to gadolinium-based contrast agents (GBCAs) and the risk of parkinsonism according to the GBCA type. Materials and Methods: Individuals aged ≥40 years who underwent first-ever magnetic resonance imaging (MRI) examinations between 2011 and 2014 were identified from the Korean nationwide population-based health insurance claims database and followed up until 2022. Individuals were divided into those who underwent at least one GBCA-enhanced MRI, and those who underwent only non-enhanced MRI. GBCA-exposed individuals were further categorized into those exposed only to linear or macrocyclic GBCAs, after excluding those exposed to both types. The primary event of interest was allcause parkinsonism. Secondary events included all-cause parkinsonism requiring medication, Parkinson’s disease (PD), atypical parkinsonism, and secondary parkinsonism. Hazard ratios (HRs) were estimated using multivariable Cox proportional hazard regression models for exposure to linear and macrocyclic GBCAs, with the non-enhanced MRI group serving as a reference. The models were adjusted for age, sex, smoking status, alcohol consumption, regular exercise, body mass index, estimated glomerular filtration rate, and comorbidities. Subgroup analyses were performed according to age, sex, renal function, and history of cancer. Results: A total of 222,977 individuals were included in this study. Among them, 92,230, 48,335, and 82,412 individuals underwent non-enhanced, linear GBCA-enhanced, and macrocyclic GBCA-enhanced MRI, respectively. Exposure to linear GBCAs slightly increased the risk of all-cause parkinsonism (adjusted HR, 1.13 [97.5% confidence interval, 1.08–1.19]), while exposure to macrocyclic GBCAs did not increase the risk (adjusted HR, 1.00 [97.5% confidence interval, 0.95–1.05]).The results were similar for all-cause parkinsonism requiring medication, PD, and secondary parkinsonism, whereas no significant association was observed for atypical parkinsonism. Conclusion Exposure to linear GBCAs may slightly increase the risk of parkinsonism in adults, whereas exposure to macrocyclic GBCAs may not. Caution should be exercised when using linear GBCAs until further evidence emerges.

Purpose: To evaluate the feasibility and utility of deep learning-based computer-aided diagnosis (DL-CAD) for the detection of pulmonary nodules on coronary artery calcium (CAC)-scoring computed tomography (CT). Materials and Methods: This retrospective study included 273 patients (aged 63.9±13.2 years; 129 men) who underwent CACscoring CT. A DL-CAD system based on thin-section images was used for pulmonary nodule detection, and two independent junior readers reviewed the standard CAC-scoring CT scans with and without referencing DL-CAD results. A reference standard was established through the consensus of two experienced radiologists. Sensitivity, positive predictive value, and F1-score were assessed on a per-nodule and per-patient basis. The patients’ medical records were monitored until November 2023. Results: A total of 269 nodules were identified in 129 patients. With DL-CAD assistance, the readers’ sensitivity significantly improved (65% vs. 80% for reader 1; 82% vs. 86% for reader 2; all p<0.001), without a notable increase in the number of false-positives per case (0.11 vs. 0.13, p=0.078 for reader 1; 0.11 vs. 0.11, p>0.999 for reader 2). Per-patient analysis also enhanced sensitivity with DL-CAD assistance (73% vs. 84%, p<0.001 for reader 1; 89% vs. 91%, p=0.250 for reader 2). During follow-up, lung cancer was diagnosed in four patients (1.5%). Among them, two had lesions detected on CAC-scoring CT, both of which were successfully identified by DL-CAD. Conclusion DL-CAD based on thin-section images can assist less experienced readers in detecting pulmonary nodules on CACscoring CT scans, improving detection sensitivity without significantly increasing false-positives.

Purpose: Renal allograft rejection, either acute or chronic, is prevalent among many recipients. This study aimed to identify multiple Doppler ultrasound parameters for predicting renal allograft rejection. Materials and Methods: Between November 2021 and April 2022, 61 renal allograft recipients were studied prospectively after excluding two patients with dual transplants and seven with hydronephrosis. The analysis excluded 11 cases (10 due to missing Doppler data or pathology reports and one due to a high interquartile range/median dispersion value), resulting in a final analysis of 50 patients. Clinical characteristics, color Doppler imaging, superb microvascular imaging, and shear-wave imaging parameters were assessed by three experienced genitourinary radiologists. The Banff classification of the biopsy tissue served as the reference standard. Univariable and multivariable logistic regression, contingency matrices, and multiple machine-learning models were employed to estimate the associations. Results: Fifty kidney transplant recipients (mean age, 53.26±8.86 years; 29 men) were evaluated. Elasticity (≤14.8 kPa) demonstrated significant associations for predicting the combination of (borderline) T cell-mediated rejection (TCMR) categories (Banff categories 3 and 4) (p=0.006) and yielded equal or higher area under the receiver operating characteristics curve (AUC) values compared to various classifiers. Dispersion (>15.0 m/s/kHz) was the only significant factor for predicting the combination of nonTCMR categories (Banff categories 2, 5, and 6) (p=0.026) and showed equal or higher AUC values than multiple machine learning classifiers. Conclusion Elasticity (≤14.8 kPa) showed a significant association with the combination of (borderline) TCMR categories, whereas dispersion (>15.0 m/s/kHz) was significantly associated with the combination of non-TCMR categories in renal allografts.

Purpose: To evaluate the feasibility and utility of deep learning-based computer-aided diagnosis (DL-CAD) for the detection of pulmonary nodules on coronary artery calcium (CAC)-scoring computed tomography (CT). Materials and Methods: This retrospective study included 273 patients (aged 63.9±13.2 years; 129 men) who underwent CACscoring CT. A DL-CAD system based on thin-section images was used for pulmonary nodule detection, and two independent junior readers reviewed the standard CAC-scoring CT scans with and without referencing DL-CAD results. A reference standard was established through the consensus of two experienced radiologists. Sensitivity, positive predictive value, and F1-score were assessed on a per-nodule and per-patient basis. The patients’ medical records were monitored until November 2023. Results: A total of 269 nodules were identified in 129 patients. With DL-CAD assistance, the readers’ sensitivity significantly improved (65% vs. 80% for reader 1; 82% vs. 86% for reader 2; all p<0.001), without a notable increase in the number of false-positives per case (0.11 vs. 0.13, p=0.078 for reader 1; 0.11 vs. 0.11, p>0.999 for reader 2). Per-patient analysis also enhanced sensitivity with DL-CAD assistance (73% vs. 84%, p<0.001 for reader 1; 89% vs. 91%, p=0.250 for reader 2). During follow-up, lung cancer was diagnosed in four patients (1.5%). Among them, two had lesions detected on CAC-scoring CT, both of which were successfully identified by DL-CAD. Conclusion DL-CAD based on thin-section images can assist less experienced readers in detecting pulmonary nodules on CACscoring CT scans, improving detection sensitivity without significantly increasing false-positives.

Purpose: Renal allograft rejection, either acute or chronic, is prevalent among many recipients. This study aimed to identify multiple Doppler ultrasound parameters for predicting renal allograft rejection. Materials and Methods: Between November 2021 and April 2022, 61 renal allograft recipients were studied prospectively after excluding two patients with dual transplants and seven with hydronephrosis. The analysis excluded 11 cases (10 due to missing Doppler data or pathology reports and one due to a high interquartile range/median dispersion value), resulting in a final analysis of 50 patients. Clinical characteristics, color Doppler imaging, superb microvascular imaging, and shear-wave imaging parameters were assessed by three experienced genitourinary radiologists. The Banff classification of the biopsy tissue served as the reference standard. Univariable and multivariable logistic regression, contingency matrices, and multiple machine-learning models were employed to estimate the associations. Results: Fifty kidney transplant recipients (mean age, 53.26±8.86 years; 29 men) were evaluated. Elasticity (≤14.8 kPa) demonstrated significant associations for predicting the combination of (borderline) T cell-mediated rejection (TCMR) categories (Banff categories 3 and 4) (p=0.006) and yielded equal or higher area under the receiver operating characteristics curve (AUC) values compared to various classifiers. Dispersion (>15.0 m/s/kHz) was the only significant factor for predicting the combination of nonTCMR categories (Banff categories 2, 5, and 6) (p=0.026) and showed equal or higher AUC values than multiple machine learning classifiers. Conclusion Elasticity (≤14.8 kPa) showed a significant association with the combination of (borderline) TCMR categories, whereas dispersion (>15.0 m/s/kHz) was significantly associated with the combination of non-TCMR categories in renal allografts.

Purpose: To evaluate the feasibility and utility of deep learning-based computer-aided diagnosis (DL-CAD) for the detection of pulmonary nodules on coronary artery calcium (CAC)-scoring computed tomography (CT). Materials and Methods: This retrospective study included 273 patients (aged 63.9±13.2 years; 129 men) who underwent CACscoring CT. A DL-CAD system based on thin-section images was used for pulmonary nodule detection, and two independent junior readers reviewed the standard CAC-scoring CT scans with and without referencing DL-CAD results. A reference standard was established through the consensus of two experienced radiologists. Sensitivity, positive predictive value, and F1-score were assessed on a per-nodule and per-patient basis. The patients’ medical records were monitored until November 2023. Results: A total of 269 nodules were identified in 129 patients. With DL-CAD assistance, the readers’ sensitivity significantly improved (65% vs. 80% for reader 1; 82% vs. 86% for reader 2; all p<0.001), without a notable increase in the number of false-positives per case (0.11 vs. 0.13, p=0.078 for reader 1; 0.11 vs. 0.11, p>0.999 for reader 2). Per-patient analysis also enhanced sensitivity with DL-CAD assistance (73% vs. 84%, p<0.001 for reader 1; 89% vs. 91%, p=0.250 for reader 2). During follow-up, lung cancer was diagnosed in four patients (1.5%). Among them, two had lesions detected on CAC-scoring CT, both of which were successfully identified by DL-CAD. Conclusion DL-CAD based on thin-section images can assist less experienced readers in detecting pulmonary nodules on CACscoring CT scans, improving detection sensitivity without significantly increasing false-positives.

Purpose: Renal allograft rejection, either acute or chronic, is prevalent among many recipients. This study aimed to identify multiple Doppler ultrasound parameters for predicting renal allograft rejection. Materials and Methods: Between November 2021 and April 2022, 61 renal allograft recipients were studied prospectively after excluding two patients with dual transplants and seven with hydronephrosis. The analysis excluded 11 cases (10 due to missing Doppler data or pathology reports and one due to a high interquartile range/median dispersion value), resulting in a final analysis of 50 patients. Clinical characteristics, color Doppler imaging, superb microvascular imaging, and shear-wave imaging parameters were assessed by three experienced genitourinary radiologists. The Banff classification of the biopsy tissue served as the reference standard. Univariable and multivariable logistic regression, contingency matrices, and multiple machine-learning models were employed to estimate the associations. Results: Fifty kidney transplant recipients (mean age, 53.26±8.86 years; 29 men) were evaluated. Elasticity (≤14.8 kPa) demonstrated significant associations for predicting the combination of (borderline) T cell-mediated rejection (TCMR) categories (Banff categories 3 and 4) (p=0.006) and yielded equal or higher area under the receiver operating characteristics curve (AUC) values compared to various classifiers. Dispersion (>15.0 m/s/kHz) was the only significant factor for predicting the combination of nonTCMR categories (Banff categories 2, 5, and 6) (p=0.026) and showed equal or higher AUC values than multiple machine learning classifiers. Conclusion Elasticity (≤14.8 kPa) showed a significant association with the combination of (borderline) TCMR categories, whereas dispersion (>15.0 m/s/kHz) was significantly associated with the combination of non-TCMR categories in renal allografts.

Purpose: To evaluate the feasibility and utility of deep learning-based computer-aided diagnosis (DL-CAD) for the detection of pulmonary nodules on coronary artery calcium (CAC)-scoring computed tomography (CT). Materials and Methods: This retrospective study included 273 patients (aged 63.9±13.2 years; 129 men) who underwent CACscoring CT. A DL-CAD system based on thin-section images was used for pulmonary nodule detection, and two independent junior readers reviewed the standard CAC-scoring CT scans with and without referencing DL-CAD results. A reference standard was established through the consensus of two experienced radiologists. Sensitivity, positive predictive value, and F1-score were assessed on a per-nodule and per-patient basis. The patients’ medical records were monitored until November 2023. Results: A total of 269 nodules were identified in 129 patients. With DL-CAD assistance, the readers’ sensitivity significantly improved (65% vs. 80% for reader 1; 82% vs. 86% for reader 2; all p<0.001), without a notable increase in the number of false-positives per case (0.11 vs. 0.13, p=0.078 for reader 1; 0.11 vs. 0.11, p>0.999 for reader 2). Per-patient analysis also enhanced sensitivity with DL-CAD assistance (73% vs. 84%, p<0.001 for reader 1; 89% vs. 91%, p=0.250 for reader 2). During follow-up, lung cancer was diagnosed in four patients (1.5%). Among them, two had lesions detected on CAC-scoring CT, both of which were successfully identified by DL-CAD. Conclusion DL-CAD based on thin-section images can assist less experienced readers in detecting pulmonary nodules on CACscoring CT scans, improving detection sensitivity without significantly increasing false-positives.

Purpose: Renal allograft rejection, either acute or chronic, is prevalent among many recipients. This study aimed to identify multiple Doppler ultrasound parameters for predicting renal allograft rejection. Materials and Methods: Between November 2021 and April 2022, 61 renal allograft recipients were studied prospectively after excluding two patients with dual transplants and seven with hydronephrosis. The analysis excluded 11 cases (10 due to missing Doppler data or pathology reports and one due to a high interquartile range/median dispersion value), resulting in a final analysis of 50 patients. Clinical characteristics, color Doppler imaging, superb microvascular imaging, and shear-wave imaging parameters were assessed by three experienced genitourinary radiologists. The Banff classification of the biopsy tissue served as the reference standard. Univariable and multivariable logistic regression, contingency matrices, and multiple machine-learning models were employed to estimate the associations. Results: Fifty kidney transplant recipients (mean age, 53.26±8.86 years; 29 men) were evaluated. Elasticity (≤14.8 kPa) demonstrated significant associations for predicting the combination of (borderline) T cell-mediated rejection (TCMR) categories (Banff categories 3 and 4) (p=0.006) and yielded equal or higher area under the receiver operating characteristics curve (AUC) values compared to various classifiers. Dispersion (>15.0 m/s/kHz) was the only significant factor for predicting the combination of nonTCMR categories (Banff categories 2, 5, and 6) (p=0.026) and showed equal or higher AUC values than multiple machine learning classifiers. Conclusion Elasticity (≤14.8 kPa) showed a significant association with the combination of (borderline) TCMR categories, whereas dispersion (>15.0 m/s/kHz) was significantly associated with the combination of non-TCMR categories in renal allografts.