Improving Automated Hemorrhage Detection at Sparse-View CT via U-Net–based Artifact Reduction

• DOI: 10.1148/ryai.230275
• Publication Date: 2024-05-08T13:51:43Z
• Authors: Johannes Thalhammer; Manuel Schultheiß; Tina Dorosti; Tobias Lasser; Franz Pfeiffer; Daniela Pfeiffer; Florian Schaff
• Keywords: Artifact (error)

Purpose To explore the potential benefits of deep learning-based artifact reduction in sparse-view cranial CT scans and its impact on automated hemorrhage detection. Materials Methods In this retrospective study, a U-Net was trained for simulated 3000 patients, obtained from public dataset reconstructed with varying levels. Additionally, EfficientNet-B2 full-view data 17 545 patients Detection performance evaluated using area under receiver operating characteristic curve (AUC), differences assessed DeLong test, along confusion matrices. A total variation (TV) postprocessing approach, commonly applied to CT, served as basis comparison. Bonferroni-corrected significance level .001/6 = .00017 used accommodate multiple hypotheses testing. Results Images were better than unprocessed TV-processed images respect image quality With postprocessing, number views could be reduced 4096 (AUC: 0.97 [95% CI: 0.97, 0.98]) 512 (0.97 0.98],