Purpose or Learning Objective To perform a comprehensive interindividual objective and subjective image quality evaluation of abdominal computed tomography (CT) images reconstructed with deep learning image reconstruction (DLIR) and hybrid iterative reconstruction (ASiR-V). Methods or Background Consecutive patients undergoing abdominal contrast-enhanced CT were prospectively enrolled from August to September 2021. Exclusion criteria were: contraindication to CT and severe motion artifacts. Thirteen datasets were reconstructed for each patient: DLIR at three strength levels (DLIR_L, DLIR_M, and DLIR_H, respectively) and ASiR-V from 10% to 100% in 10%-increments. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated to assess object image quality. Subjective image quality was assessed with a 5-point Likert scale by two independent readers. ANOVA and Kruskal-Wallis H tests were used for statistical comparison, inter-reader agreement was calculated by means of k-statistics. Post-hoc pairwise comparisons were adjusted for multiple comparisons by the Bonferroni correction. Results or Findings Fifty patients were enrolled (39 male, mean age 67±13 years). DLIR algorithm did not impact vascular attenuation compared with ASiR-V (P > 0.05). DLIR_H showed the lowest noise, comparable with ASiR-V 100% (P = 1) and significantly lower than every other reconstruction (P < 0.05). DLIR_H achieved the highest objective image quality, with SNR and CNR comparable with ASiR-V 100% (P < 0.05). DLIR_H also achieved the highest median overall subjective image quality (score 5; IQR: 4-5) with excellent inter-reader agreement (k = 0.81), comparable with DLIR_M and significantly higher than every ASIR-V dataset. Conclusion DLIR significantly improves abdominal CT image quality compared to ASiR-V, potentially improving image reconstructions in routine clinical practice.
Deep learning image reconstruction algorithm improves image quality of abdominal computed tomography: extensive comparison with hybrid iterative reconstruction / Del Gaudio, A; Guido, G; Ubaldi, N; Valanzuolo, D; Pugliese, D; Bona, G; De Santis, D; Caruso, D; Laghi, A. - (2022). (Intervento presentato al convegno 2022 European Congress of Radiology tenutosi a Wien, Austria).
Deep learning image reconstruction algorithm improves image quality of abdominal computed tomography: extensive comparison with hybrid iterative reconstruction
Del Gaudio A;Guido G;Ubaldi N;Valanzuolo D;Pugliese D;Bona G;De Santis D;Caruso D;Laghi A
2022
Abstract
Purpose or Learning Objective To perform a comprehensive interindividual objective and subjective image quality evaluation of abdominal computed tomography (CT) images reconstructed with deep learning image reconstruction (DLIR) and hybrid iterative reconstruction (ASiR-V). Methods or Background Consecutive patients undergoing abdominal contrast-enhanced CT were prospectively enrolled from August to September 2021. Exclusion criteria were: contraindication to CT and severe motion artifacts. Thirteen datasets were reconstructed for each patient: DLIR at three strength levels (DLIR_L, DLIR_M, and DLIR_H, respectively) and ASiR-V from 10% to 100% in 10%-increments. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated to assess object image quality. Subjective image quality was assessed with a 5-point Likert scale by two independent readers. ANOVA and Kruskal-Wallis H tests were used for statistical comparison, inter-reader agreement was calculated by means of k-statistics. Post-hoc pairwise comparisons were adjusted for multiple comparisons by the Bonferroni correction. Results or Findings Fifty patients were enrolled (39 male, mean age 67±13 years). DLIR algorithm did not impact vascular attenuation compared with ASiR-V (P > 0.05). DLIR_H showed the lowest noise, comparable with ASiR-V 100% (P = 1) and significantly lower than every other reconstruction (P < 0.05). DLIR_H achieved the highest objective image quality, with SNR and CNR comparable with ASiR-V 100% (P < 0.05). DLIR_H also achieved the highest median overall subjective image quality (score 5; IQR: 4-5) with excellent inter-reader agreement (k = 0.81), comparable with DLIR_M and significantly higher than every ASIR-V dataset. Conclusion DLIR significantly improves abdominal CT image quality compared to ASiR-V, potentially improving image reconstructions in routine clinical practice.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
