Advancing radiological workflows through AI: Deep learning for automated tissue quantification, disease classification, generating synthetic contrast imaging and free-text report content extraction

Abstract

<strong>The following publications are included in this cumulative habilitation thesis</strong><br/> Advances in artificial intelligence (AI) algorithms have raised expectations for the transformation of knowledge-based workflows, also in radiology. In this thesis, the applicability of AI to automate, optimize or support various image- or report-based analysis was investigated. It was the aim to provide insights into the potential of AI for advancing radiological workflows and thereby improving patient care. The scope of the eight original works included in this cumulative thesis can be categorized into three main topics:<br/> <span style="text-decoration: underline;">Processing of free-text radiological reports</span><br/> 1. <strong>Privacy-ensuring, open-weights large language models are competitive with closed GPT-4o in extracting chest X-ray findings from free-text reports.</strong> Nowak S, Wulff B, Layer YC, Theis M, Isaak A, Salam B, Block W, Kuetting D, Pieper CCC, Luetkens JA, Attenberger UI, Sprinkart AM. Radiology. 2024; in press.<br/> 2. <strong>Transformer-based structuring of free-text radiology report databases.</strong> Nowak S*, Biesner D*, Layer Y, Theis M, Schneider H, Block W, Wulff B, Attenberger UI*, Sifa R*, Sprinkart AM*. European Radiology. 2023;33(6):4228–4236.<br/> 3. <strong>Development of image-based decision support systems utilizing information extracted from radiological free-text report databases with text-based transformers.</strong> Nowak S*, Schneider H*, Layer YC, Theis M, Biesner D, Block W, Wulff B, Attenberger UI, Sifa R*, Sprinkart AM*. European Radiology. 2024;34(5):2895-2904.<br/> <span style="text-decoration: underline;">Generating synthetic radiological images</span><br/> 4. <strong>Deep learning virtual contrast-enhanced T1 mapping for contrast-free myocardial extracellular volume assessment. Journal of the American Heart Association.</strong> Nowak S*, Bischoff LM*, Pennig L, Kaya K, Isaak A, Theis M, Block W, Pieper CC, Kuetting D, Zimmer S, Nickenig G, Attenberger UI, Sprinkart AM*, Luetkens JA*. Journal of the American Heart Association. 2024;13(19):e035599. <br/> <span style="text-decoration: underline;">Supporting diagnostic and treatment decisions based on radiological imaging</span><br/> 5. <strong>Deep learning supports the differentiation of alcoholic and other-than-alcoholic cirrhosis based on MRI.</strong> Luetkens JA*, Nowak S*, Mesropyan N, Block W, Praktiknjo M, Chang J, Bauckhage C, Sifa R, Sprinkart AM*, Faron A*, Attenberger UI*. Scientific reports. 2022;12(1):8297.<br/> 6. <strong>Deep learning–based assessment of CT markers of sarcopenia and myosteatosis for outcome assessment in patients with advanced pancreatic cancer after high-intensity focused ultrasound treatment.</strong> Nowak S*, Kloth C*, Theis M, Marinova M, Attenberger UI, Sprinkart AM*, Luetkens JA*. European Radiology. 2024;34(1):279–286.<br/> 7. <strong>Direct deep learning-based survival prediction from pre-interventional CT prior to transcatheter aortic valve replacement.</strong> Theis M, Block W, Luetkens JA, Attenberger UI, Nowak S*, Sprinkart AM*. European Journal of Radiology. 2023;168:111150.<br/> 8. <strong>Computer tomography-based assessment of perivascular adipose tissue in patients with abdominal aortic aneurysms.</strong> Ginzburg D*, Nowak S*, Attenberger U, Luetkens J, Sprinkart AM*, Kuetting D*. Scientific Reports 2024;14(1):20512.<br/> <br/> * contributed equally