Improving Automated Hemorrhage Detection in Sparse-view Computed Tomography via Deep Convolutional Neural Network based Artifact Reduction

• DOI: 10.48550/arxiv.2303.09340
• Publication Date: 2023-01-01
• Authors: Johannes Thalhammer; Manuel Schultheiß; Tina Dorosti; Tobias Lasser; Franz Pfeiffer; Daniela Pfeiffer; Florian Schaff
• Keywords: Artifact (error); Confusion

This is a preprint. The latest version has been published here: https://pubs.rsna.org/doi/10.1148/ryai.230275 Purpose: Sparse-view computed tomography (CT) an effective way to reduce dose by lowering the total number of views acquired, albeit at expense image quality, which, in turn, can impact ability detect diseases. We explore deep learning-based artifact reduction sparse-view cranial CT scans and its on automated hemorrhage detection. Methods: trained U-Net for artefact simulated from 3000 patients obtained public dataset reconstructed with varying levels sub-sampling. Additionally, we convolutional neural network fully sampled data 17,545 evaluated classification performance using area under receiver operator characteristic curves (AUC-ROCs) corresponding 95% confidence intervals (CIs) DeLong test, along confusion matrices. was compared analytical approach based variation (TV). Results: performed superior unprocessed TV-processed images respect quality diagnosis. With post-processing, be reduced 4096 (AUC-ROC: 0.974; CI: 0.972-0.976) 512 (0.973; 0.971-0.975) minimal decrease detection (P<.001) 256 (0.967; 0.964-0.969) slight (P<.001). Conclusion: results suggest that substantially enhances CTs. Our findings highlight appropriate post-processing crucial optimal diagnostic accuracy while minimizing radiation dose.