To examine the comparative robustness of computed tomography (CT)-based conventional radiomics and deep-learning convolutional neural networks (CNN) to predict overall survival (OS) in HCC patients. Retrospectively, 114 HCC patients with pretherapeutic CT of the liver were randomized into a development (n = 85) and a validation (n = 29) cohort, including patients of all tumor stages and several applied therapies. In addition to clinical parameters, image annotations of the liver parenchyma and of tumor findings on CT were available. Cox-regression based on radiomics features and CNN models were established and combined with clinical parameters to predict OS. Model performance was assessed using the concordance index (C-index). Log-rank tests were used to test model-based patient stratification into high/low-risk groups. The clinical Cox-regression model achieved the best validation performance for OS (C-index [95% confidence interval (CI)] 0.74 [0.57–0.86]) with a significant difference between the risk groups (p = 0.03). In image analysis, the CNN models (lowest C-index [CI] 0.63 [0.39–0.83]; highest C-index [CI] 0.71 [0.49–0.88]) were superior to the corresponding radiomics models (lowest C-index [CI] 0.51 [0.30–0.73]; highest C-index [CI] 0.66 [0.48–0.79]). A significant risk stratification was not possible (p > 0.05). Under clinical conditions, CNN-algorithms demonstrate superior prognostic potential to predict OS in HCC patients compared to conventional radiomics approaches and could therefore provide important information in the clinical setting, especially when clinical data is limited.

Radiological reporting errors have a direct negative impact on patient treatment. The purpose of this study was to investigate the contribution of clinical information (CI) in radiological reporting of oncological imaging and the dependence on the radiologists’ experience level (EL). Sixty-four patients with several types of carcinomas and twenty patients without tumors were enrolled. Computed tomography datasets acquired in primary or follow-up staging were independently analyzed by three radiologists (R) with different EL (R1: 15 years; R2: 10 years, R3: 1 year). Reading was initially performed without and 3 months later with CI. Overall, diagnostic accuracy and sensitivity for primary tumor detection increased significantly when receiving CI from 77% to 87%; p = 0.01 and 73% to 83%; p = 0.01, respectively. All radiologists benefitted from CI; R1: 85% vs. 92%, p = 0.15; R2: 77% vs. 83%, p = 0.33; R3: 70% vs. 86%, p = 0.02. Overall, diagnostic accuracy and sensitivity for detecting lymphogenous metastases increased from 80% to 85% (p = 0.13) and 42% to 56% (p = 0.13), for detection of hematogenous metastases from 85% to 86% (p = 0.61) and 46% to 60% (p = 0.15). Specificity remained stable (>90%). Thus, CI in oncological imaging seems to be essential for correct radiological reporting, especially for residents, and should be available for the radiologist whenever possible.

Background: The integration of artificial intelligence (AI) into radiology aims to improve diagnostic accuracy and efficiency, particularly in settings with limited access to expert radiologists and in times of personnel shortage. However, challenges such as insufficient validation in actual real-world settings or automation bias should be addressed before implementing AI software in clinical routine. Methods: This cross-sectional study in a maximum care hospital assesses the concordance between diagnoses made by a commercial AI-based software and conventional radiological methods augmented by AI for four major thoracic pathologies in chest X-ray: fracture, pleural effusion, pulmonary nodule and pneumonia. Chest radiographs of 1506 patients (median age 66 years, 56.5% men) consecutively obtained between January and August 2023 were re-evaluated by the AI software InferRead DR Chest^®. Results: Overall, AI software detected thoracic pathologies more often than radiologists (18.5% vs. 11.1%). In detail, it detected fractures, pneumonia, and nodules more frequently than radiologists, while radiologists identified pleural effusions more often. Reliability was highest for pleural effusions (0.63, 95%-CI 0.58–0.69), indicating good agreement, and lowest for fractures (0.39, 95%-CI 0.32–0.45), indicating moderate agreement. Conclusions: The tested software shows a high detection rate, particularly for fractures, pneumonia, and nodules, but hereby produces a nonnegligible number of false positives. Thus, AI-based software shows promise in enhancing diagnostic accuracy; however, cautious interpretation and human oversight remain crucial.

To investigate factors influencing the success and complication rate of percutaneous transhepatic biliary drainage (PTBD). PTBD procedures between 2017 and 2022 were enrolled retrospectively. Success rate, complications, and technical considerations were compared using Mann-Whitney U-, X²-, and Fishers exact tests. The influence of the study center's experience (training-effect) on success and complication rates was assessed by linear regression categorized by year. In 398 patients, 499 PTBD procedures (337 men; mean age 66.2 +/- 12.5 years) were carried out, with a success rate of 83.2% (415/499). PTBD in dilated bile ducts was more successful than in non-dilated bile ducts (90.0%; 316/340 vs. 68.6%; 109/159; p < 0.001), with e.g. lower radiation doses (2787.52 +/- 4012.72 cGy*cm² vs. 4679.25 +/- 4663.55 cGy*cm²; p < 0.001), and shorter total procedure time (33.42 +/- 24.03 min vs. 41.09 +/- 27.21 min; p < 0.001). Complications occurred in 34/499 (6.8%) procedures (major complications n = 25/34) with no significant difference in bile duct width. Right-sided PTBD revealed more complications (9.0%; 30/332 vs. 2.4%; 4/166; p = 0.006) and higher radiation doses (3679.47 +/- 4571.71 cGy*cm2 vs. 2819.01 +/- 3724.92 cGy*cm2; p = 0.001) than left-sided approaches. Linear regression showed a significant continuous increase in the technical success rate of 3.0% per year (2017-2022; 72.5%; 78.5%; 82.2%; 85.0%; 89.0%; 87.5%; p = 0.005), while the overall complication rate remained unaffected (p = 0.364). Medical centers adopting PTBD procedures can potentially increase their success rate significantly within a short period of time. PTBD is a safe procedure, with left-sided approaches showing lower complication rates and radiation exposure, underscoring their often-underestimated advantages in clinical practice.

Purpose: To develop and validate a vendor‐agnostic, motion‐insensitive proton‐density fat‐fraction (PDFF) quantification method. Methods: Flip‐angle‐modulated (FAM) 2D chemical‐shift‐encoded (CSE) MRI for PDFF quantification was implemented in both the vendor‐agnostic platform Pulseq (“Pulseq‐FAM”) and one vendor‐specific platform (“GE‐specific FAM”). These implementations were distributed to four sites with twelve MR systems of three vendors (Siemens/GE/Philips) and field strengths (0.55T/1.5T/3T). A sequentially‐shipped 16‐vial phantom (PDFF = 0%–30%/T1 _water = 200–1400 ms) underwent confounder‐corrected PDFF mapping with commercial 3D‐CSE methods and GE‐specific FAM as available on each system, and Pulseq‐FAM on every system. To assess bias, phantom PDFF measurements were compared to reference. Between‐system variance was evaluated with linear mixed‐effects modeling. Different volunteers were also imaged at each site to assess free‐breathing PDFF mapping feasibility. A prospective single‐site volunteer study was also conducted. Adult patients and children were imaged with breath‐held 3D‐CSE and free‐breathing GE‐specific and Pulseq‐FAM. Radiologists evaluated images for overall quality and motion artifacts. To assess bias, Pulseq‐FAM PDFF measurements were compared to 3D‐CSE and GE‐specific FAM. Test–retest repeatability was assessed by re‐imaging after repositioning. Between‐field‐strength reproducibility was assessed at 1.5T and 3.0T. Results: In the multi‐center study, Pulseq‐FAM showed reduced T1‐bias and between‐system variability versus 3D‐CSE in phantom PDFF measurements, and free‐breathing feasibility in volunteers. In the single‐site volunteer study ( N = 57), Pulseq‐FAM improved image quality and motion artifacts versus 3D‐CSE ( p < 0.01). Pulseq‐FAM showed excellent agreement with 3D‐CSE (95% limits‐of‐agreement (LoA) = 3.4% PDFF) and GE‐specific FAM (LoA = 2.0%). Pulseq‐FAM showed excellent repeatability (repeatability coefficient (RC) = 1.6% PDFF) and between‐field‐strength reproducibility (reproducibility coefficient (RDC) = 2.4%) versus 3D‐CSE (RC = 2.7%/RDC = 3.4%; differences p < 0.05). Conclusion: Pulseq‐FAM enables accurate, reproducible, vendor‐agnostic, and motion‐insensitive PDFF quantification in adults and children.

Background: The risk of postoperative pancreatic fistula (POPF), one of the most dreaded complications after pancreatic surgery, can be predicted from preoperative imaging and tabular clinical routine data. However, existing studies suffer from limited clinical applicability due to a need for manual data annotation and a lack of external validation. We propose AutoFRS (automated fistula risk score software), an externally validated end-to-end prediction tool for POPF risk stratification based on multimodal preoperative data. Materials and methods: We trained AutoFRS on preoperative contrast-enhanced computed tomography imaging and clinical data from 108 patients undergoing pancreatic head resection and validated it on an external cohort of 61 patients. Prediction performance was assessed using the area under the receiver operating characteristic curve (AUC) and balanced accuracy. In addition, model performance was compared to the updated alternative fistula risk score (ua-FRS), the current clinical gold standard method for intraoperative POPF risk stratification. Results: AutoFRS achieved an AUC of 0.81 and a balanced accuracy of 0.72 in internal validation and an AUC of 0.79 and a balanced accuracy of 0.70 in external validation. In a patient subset with documented intraoperative POPF risk factors, AutoFRS (AUC: 0.84 ± 0.05) performed on par with the uaFRS (AUC: 0.85 ± 0.06). The AutoFRS web application facilitates annotation-free prediction of POPF from preoperative imaging and clinical data based on the AutoFRS prediction model. Conclusion: POPF can be predicted from multimodal clinical routine data without human data annotation, automating the risk prediction process. We provide additional evidence of the clinical feasibility of preoperative POPF risk stratification and introduce a software pipeline for future prospective evaluation. Export