RESUMEN
Background: Integrating complementary diagnostic data sources promises enhanced robustness in the predictive performance of artificial intelligence (AI) models, a crucial requirement for future clinical validation/implementation. In this study, we investigate the potential value of integrating data from noninvasive diagnostic modalities, including chest computed tomography (CT) imaging, routine laboratory blood tests, and clinical parameters, to retrospectively predict 1-year survival in a cohort of patients with advanced non-small-cell lung cancer, melanoma, and urothelial cancer treated with immunotherapy. Patients and methods: The study included 475 patients, of whom 444 had longitudinal CT scans and 475 had longitudinal laboratory data. An ensemble of AI models was trained on data from each diagnostic modality, and subsequently, a model-agnostic integration approach was adopted for combining the prediction probabilities of each modality and producing an integrated decision. Results: Integrating different diagnostic data demonstrated a modest increase in predictive performance. The highest area under the curve (AUC) was achieved by CT and laboratory data integration (AUC of 0.83, 95% confidence interval 0.81-0.85, P < 0.001), whereas the performance of individual models trained on laboratory and CT data independently yielded AUCs of 0.81 and 0.73, respectively. Conclusions: In our retrospective cohort, integrating different noninvasive data modalities improved performance.
RESUMEN
BACKGROUND: To investigate whether aortic valve calcification (AVC) scoring performed with different workstation platforms generates comparable and thus software-independent results. METHODS: In this IRB-approved retrospective study, we included 100 consecutive patients with symptomatic aortic stenosis undergoing CT prior to transcatheter aortic valve implantation. Two independent observers performed AVC scoring on non-enhanced images with commercially available software platforms of four vendors (GE, Philips, Siemens, 3mensio). Gender-specific Agatston score cut-off values were applied according to current recommendations to assign patients to different likelihood categories of aortic stenosis (unlikely to very likely). Comparative analysis of Agatston scores between the four platforms were performed by using Kruskal-Wallis analysis, Spearman rank correlation, linear regression analysis, and Bland-Altman analysis. Differences in category assignment were compared using Fisher's exact test and Cohen's kappa. RESULTS: For both observers, each workstation platform produced slightly different numeric AVC Agatston scores, however, without statistical significance (pâ¯=â¯0.96 and pâ¯=â¯0.98). Excellent correlation was found between platforms, with râ¯=â¯0.991-0.996 (Spearman) and r2â¯=â¯0.981-0.992 (regression analysis) for both observers. Bland-Altman analyses revealed small mean differences with narrow limits of agreement between platforms (mean differences: 6⯱â¯128 to 100⯱â¯179), for inter-observer (mean differences: 1⯱â¯43 to 12⯱â¯70), and intra-observer variability (mean differences: 9⯱â¯42 to 20⯱â¯96). Observer 1 assigned 11 (kappa: 0.85-0.97) and observer 2 assigned 10 patients (kappa: 0.88-0.95) to different likelihood groups of severe aortic stenosis with at least one platform. Overall, there was no significant difference of likelihood assignment between platforms (pâ¯=â¯0.98 and pâ¯=â¯1.0, respectively). CONCLUSION: While absolute values differ slightly, common commercially available software platforms produce comparable results for AVC scoring, which indicates software-independence of the method.
Asunto(s)
Estenosis de la Válvula Aórtica/diagnóstico por imagen , Válvula Aórtica/diagnóstico por imagen , Válvula Aórtica/patología , Calcinosis/diagnóstico por imagen , Interpretación de Imagen Radiográfica Asistida por Computador/métodos , Programas Informáticos , Tomografía Computarizada por Rayos X , Anciano , Anciano de 80 o más Años , Femenino , Humanos , Masculino , Variaciones Dependientes del Observador , Valor Predictivo de las Pruebas , Reproducibilidad de los Resultados , Estudios Retrospectivos , Índice de Severidad de la EnfermedadRESUMEN
PURPOSE: To compare prospectively, in patients undergoing chest computed tomography (CT) for pulmonary-nodules or infection, image-quality and accuracy of standard dose (SD) and reduced dose (RD) CT with tin-filtration. MATERIAL AND METHODS: This IRB-approved study included 100 consecutive patients (36 female;median age 56 years) referred for follow-up of pulmonary-nodules (n=43) or suspicion of infection (n=57) undergoing single-energy CT with SD and RD using tin-filtration at 100 kVp (CTDIvol 2.47 mGy and 0.07 mGy, respectively). Images were reconstructed with advanced modeled iterative reconstruction (ADMIRE) at strength 3 and 5. Image-noise was measured. Two independent readers evaluated nodules and pulmonary-infection. SD CT served as reference standard. RESULTS: No significant difference was found in noise between RD with ADMIRE5 and SD with ADMIRE3 (118HU ± 14 vs. 120HU ± 17; p=0.08). Sensitivity for detection of atelectasis and interstitial lung changes was higher in images reconstructed with ADMIRE5 (93% and 88%; respectively) than in those reconstructed with ADIMRE3 (77% and 78%; respectively). Sensitivity for detection of consolidations was 90% for ADMIRE3 and 89% for ADMIRE5. Sensitivity for nodule detection was 71% for ADMIRE3 and 81% for ADMIRE5. Specificity for detection of atelectasis and interstitial lung changes was 99% and 96% with ADMIRE5 and 99% and 96% with ADMIRE3. Specificity for detection of consolidations was 99% for ADMIRE3 and 5. Specificity for detection of nodules was 87% for both ADMIRE3 and 5. CONCLUSION: Chest CT with a radiation dose equivalent to conventional radiography is feasible and allows for detection of pulmonary infection with high sensitivity, whereas the accuracy for detecting nodules is only moderate.