Development of an image-based Random Forest classifier for prediction of surgery duration of laparoscopic sigmoid resections.

PURPOSE: Sigmoid diverticulitis is a disease with a high socioeconomic burden, accounting for a high number of left-sided colonic resections worldwide. Modern surgical scheduling relies on accurate prediction of operation times to enhance patient care and optimize healthcare resources. This study aims to develop a predictive model for surgery duration in laparoscopic sigmoid resections, based on preoperative CT biometric and demographic patient data. METHODS: This retrospective single-center cohort study included 85 patients who underwent laparoscopic sigmoid resection for diverticular disease. Potentially relevant procedure-specific anatomical parameters recommended by a surgical expert were measured in preoperative CT imaging. After random split into training and test set (75% / 25%) multiclass logistic regression was performed and a Random Forest classifier was trained on CT imaging parameters, patient age, and sex in the training cohort to predict categorized surgery duration. The models were evaluated in the test cohort using established performance metrics including receiver operating characteristics area under the curve (AUROC). RESULTS: The Random Forest model achieved a good average AUROC of 0.78. It allowed a very good prediction of long (AUROC = 0.89; specificity 0.71; sensitivity 1.0) and short (AUROC = 0.81; specificity 0.77; sensitivity 0.56) procedures. It clearly outperformed the multiclass logistic regression model (AUROC: average = 0.33; short = 0.31; long = 0.22). CONCLUSION: A Random Forest classifier trained on demographic and CT imaging biometric patient data could predict procedure duration outliers of laparoscopic sigmoid resections. Pending validation in a multicenter study, this approach could potentially improve procedure scheduling in visceral surgery and be scaled to other procedures.

Algorithmic transparency and interpretability measures improve radiologists' performance in BI-RADS 4 classification.

OBJECTIVE: To evaluate the perception of different types of AI-based assistance and the interaction of radiologists with the algorithm's predictions and certainty measures. METHODS: In this retrospective observer study, four radiologists were asked to classify Breast Imaging-Reporting and Data System 4 (BI-RADS4) lesions (n = 101 benign, n = 99 malignant). The effect of different types of AI-based assistance (occlusion-based interpretability map, classification, and certainty) on the radiologists' performance (sensitivity, specificity, questionnaire) were measured. The influence of the Big Five personality traits was analyzed using the Pearson correlation. RESULTS: Diagnostic accuracy was significantly improved by AI-based assistance (an increase of 2.8% ± 2.3%, 95 %-CI 1.5 to 4.0 %, p = 0.045) and trust in the algorithm was generated primarily by the certainty of the prediction (100% of participants). Different human-AI interactions were observed ranging from nearly no interaction to humanization of the algorithm. High scores in neuroticism were correlated with higher persuasibility (Pearson's r = 0.98, p = 0.02), while higher consciousness and change of accuracy showed an inverse correlation (Pearson's r = -0.96, p = 0.04). CONCLUSION: Trust in the algorithm's performance was mostly dependent on the certainty of the predictions in combination with a plausible heatmap. Human-AI interaction varied widely and was influenced by personality traits. KEY POINTS: • AI-based assistance significantly improved the diagnostic accuracy of radiologists in classifying BI-RADS 4 mammography lesions. • Trust in the algorithm's performance was mostly dependent on the certainty of the prediction in combination with a reasonable heatmap. • Personality traits seem to influence human-AI collaboration. Radiologists with specific personality traits were more likely to change their classification according to the algorithm's prediction than others.

Evaluation of GPT-4's Chest X-Ray Impression Generation: A Reader Study on Performance and Perception.

Exploring the generative capabilities of the multimodal GPT-4, our study uncovered significant differences between radiological assessments and automatic evaluation metrics for chest x-ray impression generation and revealed radiological bias.

Additional MRI for initial M-staging in pancreatic cancer: a cost-effectiveness analysis.

OBJECTIVE: Pancreatic cancer is portrayed to become the second leading cause of cancer-related death within the next years. Potentially complicating surgical resection emphasizes the importance of an accurate TNM classification. In particular, the failure to detect features for non-resectability has profound consequences on patient outcomes and economic costs due to incorrect indication for resection. In the detection of liver metastases, contrast-enhanced MRI showed high sensitivity and specificity; however, the cost-effectiveness compared to the standard of care imaging remains unclear. The aim of this study was to analyze whether additional MRI of the liver is a cost-effective approach compared to routinely acquired contrast-enhanced computed tomography (CE-CT) in the initial staging of pancreatic cancer. METHODS: A decision model based on Markov simulation was developed to estimate the quality-adjusted life-years (QALYs) and lifetime costs of the diagnostic modalities. Model input parameters were assessed based on evidence from recent literature. The willingness-to-pay (WTP) was set to $100,000/QALY. To evaluate model uncertainty, deterministic and probabilistic sensitivity analyses were performed. RESULTS: In the base-case analysis, the model yielded a total cost of $185,597 and an effectiveness of 2.347 QALYs for CE-MR/CT and $187,601 and 2.337 QALYs for CE-CT respectively. With a net monetary benefit (NMB) of $49,133, CE-MR/CT is shown to be dominant over CE-CT with a NMB of $46,117. Deterministic and probabilistic survival analysis showed model robustness for varying input parameters. CONCLUSION: Based on our results, combined CE-MR/CT can be regarded as a cost-effective imaging strategy for the staging of pancreatic cancer. KEY POINTS: • Additional MRI of the liver for initial staging of pancreatic cancer results in lower total costs and higher effectiveness. • The economic model showed high robustness for varying input parameters.

Radiologic predictors for failure of non-operative management of complicated diverticulitis: a single-centre cohort study.

PURPOSE: Modern non-operative management of diverticulitis consists of a complex therapeutic regimen and is successful in most cases even of complicated diverticulitis. Still, a certain proportion of patients requires urgent surgery due to failure of the conservative approach. This study aims to identify predictors for failure of conservative treatment of complicated diverticulitis with the need for subsequent urgent resection during the acute episode. METHODS: A single-centre retrospective cohort study was performed at our tertiary centre including cases of acute complicated diverticulitis (characterized by localized abscess formation and/or pericolic air) between 2007 and 2019 that were treated guideline-conform by multimodal conservative treatment. Radiologic characteristics of disease in CT scans upon admission were analysed by uni- and multivariable logistic regression to determine predictors for resection within 30 days after onset of the conservative therapy approach. RESULTS: A total of 669 cases of acute diverticulitis were identified, of which 141 patients met the inclusion criteria. Overall, 13% (n = 19) of patients were operated within 30 days despite initial conservative management. Multivariable logistic regression identified length of inflamed bowel greater than 7 cm (p < 0.011) and abscess formations >1 cm (p < 0.001) as significant risk factors for failure of conservative treatment. CONCLUSION: Patients with length of inflamed bowel >7 cm or abscess formation >1 cm have increased risk for failure of conservative treatment of acute episodes of diverticulitis with contained perforations with subsequent need for urgent surgery. Therefore, conservative treatment of those patients should be monitored with special caution.

Improved detection rates and treatment planning of head and neck cancer using dual-layer spectral CT.

PURPOSE: The aim of this study was to evaluate the advantages of dual-layer spectral CT (DLSCT) in detection and staging of head and neck cancer (HNC) as well as the imaging of tumour margins and infiltration depth compared to conventional contrast enhanced CT (CECT). MATERIALS AND METHODS: Thirty-nine patients with a proven diagnosis of HNC were examined with a DLSCT scanner and retrospectively analysed. An age-matched healthy control group of the same size was used. Images were acquired in the venous phase. Virtual monoenergetic 40keV-equivalent (MonoE40) images were compared to CECT-images. Diagnostic confidence for tumour identification and margin detection was rated independently by four experienced observers. The steepness of the Hounsfield unit (HU)-increase at the tumour margin was analysed. External carotid artery branch image reconstructions were performed and their contrast compared to conventional arterial phase imaging. Means were compared using a Student's t-test. ANOVA was used for multiple comparisons. RESULTS: MonoE40 images were superior to CECT-images in tumour detection and margin delineation. MonoE40 showed significantly higher attenuation differences between tumour and healthy tissue compared to CECT-images (p < 0.001). The HU-increase at the boundary of the tumour was significantly steeper in MonoE40 images compared to CECT-images (p < 0.001). Iodine uptake in the tumour was significantly higher compared to healthy tissue (p < 0.001). MonoE40 compared to conventional images allowed visualisation of external carotid artery branches from the venous phase in a higher number of cases (87% vs. 67%). CONCLUSION: DLSCT enables improved detection of primary and recurrent head and neck cancer and quantification of tumour iodine uptake. Improved contrast of MonoE40 compared to conventional reconstructions enables higher diagnostic confidence concerning tumour margin detection and vessel identification. KEY POINTS: • Sensitivity concerning tumour detection are higher using dual-layer spectral-CT than conventional CT. • Lesion to background contrast in DLSCT is significantly higher than in CECT. • DLSCT provides sufficient contrast for evaluation of external carotid artery branches.

Evaluating Treatment Response in GEJ Adenocarcinoma: The Role of Pretherapeutic and Posttherapeutic Iodine Mapping.

BACKGROUND: Neoadjuvant therapy regimens have significantly improved the prognosis of GEJ (gastroesophageal junction) cancer; however, there are a significant percentage of patients who benefit from earlier resection or adapted therapy regimens, and the true response rate can only be determined histopathologically. Methods that allow preoperative assessment of response are lacking. PURPOSE: The purpose of this retrospective study is to assess the potential of pretherapeutic and posttherapeutic spectral CT iodine density (IoD) in predicting histopathological response to neoadjuvant chemotherapy in patients diagnosed with adenocarcinoma of the GEJ. METHODS: In this retrospective cohort study, a total of 62 patients with GEJ carcinoma were studied. Patients received a multiphasic CT scan at diagnosis and preoperatively. Iodine-density maps were generated based on spectral CT data. All tumors were histopathologically analyzed, and the tumor regression grade (TRG) according to Becker et al ( Cancer . 2003;98:1521-1530) was determined. Two experienced radiologists blindly placed 5 defined ROIs in the tumor region of highest density, and the maximum value was used for further analysis. Iodine density was normalized to the aortic iodine uptake. In addition, tumor response was assessed according to standard RECIST measurement. After assessing interrater reliability, the correlation of IoD values with treatment response and with histopathologic TRG was evaluated. RESULTS: The normalized ΔIoD (IoD at diagnosis - IoD after neoadjuvant treatment) and the normalized IoD after neoadjuvant treatment correlated significantly with the TRG. For the detection of responders and nonresponders, the receiver operating characteristic (ROC) curve for normalized ΔIoD yielded the highest area under the curve of 0.95 and achieved a sensitivity and specificity of 92.3% and 92.1%, respectively. Iodine density after neoadjuvant treatment achieved an area under the curve of 0.88 and a sensitivity and specificity of 86.8% and 84.6%, respectively (cutoff, 0.266). Iodine density at diagnosis and RECIST did not provide information to distinguish responders from nonresponders. Using the cutoff value for IoD after neoadjuvant treatment, a reliable classification of responders and nonresponders was achieved for both readers in a test set of 11 patients. Intraclass correlation coefficient revealed excellent interrater reliability (intraclass correlation coefficient, >0.9). Lastly, using the cutoff value for normalized ΔIoD as a definition for treatment response, a significantly longer survival of responders was shown. CONCLUSIONS: Changes in IoD after neoadjuvant treatment of GEJ cancer may be a potential surrogate for therapy response.

Comparison of Virtual Non-Contrast and True Non-Contrast CT Images Obtained by Dual-Layer Spectral CT in COPD Patients.

Chronic obstructive pulmonary disease (COPD) is one of the leading causes of death. Recent studies have underlined the importance of non-contrast-enhanced chest CT scans not only for emphysema progression quantification, but for correlation with clinical outcomes as well. As about 40 percent of the 300 million CT scans per year are contrast-enhanced, no proper emphysema quantification is available in a one-stop-shop approach for patients with known or newly diagnosed COPD. Since the introduction of spectral imaging (e.g., dual-energy CT scanners), it has been possible to create virtual non-contrast-enhanced images (VNC) from contrast-enhanced images, making it theoretically possible to offer proper COPD imaging despite contrast enhancing. This study is aimed towards investigating whether these VNC images are comparable to true non-contrast-enhanced images (TNC), thereby reducing the radiation exposure of patients and usage of resources in hospitals. In total, 100 COPD patients with two scans, one with (VNC) and one without contrast media (TNC), within 8 weeks or less obtained by a spectral CT using dual-layer technology, were included in this retrospective study. TNC and VNC were compared according to their voxel-density histograms. While the comparison showed significant differences in the low attenuated volumes (LAVs) of TNC and VNC regarding the emphysema threshold of -950 Houndsfield Units (HU), the 15th and 10th percentiles of the LAVs used as a proxy for pre-emphysema were comparable. Upon further investigation, the threshold-based LAVs (-950 HU) of TNC and VNC were comparable in patients with a water equivalent diameter (DW) below 270 mm. The study concludes that VNC imaging may be a viable option for assessing emphysema progression in COPD patients, particularly those with a normal body mass index (BMI). Further, pre-emphysema was generally comparable between TNC and VNC. This approach could potentially reduce radiation exposure and hospital resources by making additional TNC scans obsolete.

Spectral computed tomography angiography using a gadolinium-based contrast agent for imaging of pathologies of the aorta.

OBJECTIVES: Especially patients with aortic aneurysms and multiple computed tomography angiographies (CTA) might show medical conditions which oppose the use of iodine-based contrast agents. CTA using monoenergetic reconstructions from dual layer CT and gadolinium (Gd-)based contrast agents might be a feasible alternative in these patients. Therefore, the purpose of this study was to evaluate the feasibility of clinical spectral CTA with a Gd-based contrast agent in patients with aortic aneurysms. METHODS: Twenty-one consecutive scans in 15 patients with and without endovascular aneurysm repair showing contraindications for iodine-based contrast agents were examined using clinical routine doses (0.2 mmol/kg) of Gd-based contrast agent with spectral CT. Monoenergetic reconstructions of the spectral data set were computed. RESULTS: There was a significant increase in the intravascular attenuation of the aorta between pre- and post-contrast images for the MonoE40 images in the thoracic and the abdominal aorta (p < 0.001 for both). Additionally, the ratio between pre- and post-contrast images was significantly higher in the MonoE40 images as compared to the conventional images with a factor of 6.5 ± 4.5 vs. 2.4 ± 0.5 in the thoracic aorta (p = 0.003) and 4.1 ± 1.8 vs. 1.9 ± 0.5 in the abdominal aorta (p < 0.001). CONCLUSIONS: To conclude, our study showed that Gd-CTA is a valid and reliable alternative for diagnostic imaging of the aorta for clinical applications. Monoenergetic reconstructions of computed tomography angiographies using gadolinium based contrast agents may be a useful alternative in patients with aortic aneurysms and contraindications for iodine based contrast agents.

Machine learning assisted feature identification and prediction of hemodynamic endpoints using computed tomography in patients with CTEPH.

Chronic thromboembolic pulmonary hypertension (CTEPH) is a rare but potentially curable cause of pulmonary hypertension (PH). Currently PH is diagnosed by right heart catheterisation. Computed tomography (CT) is used for ruling out other causes and operative planning. This study aims to evaluate importance of different quantitative/qualitative imaging features and develop a supervised machine learning (ML) model to predict hemodynamic risk groups. 127 Patients with diagnosed CTEPH who received preoperative right heart catheterization and thoracic CTA examinations (39 ECG-gated; 88 non-ECG gated) were included. 19 qualitative/quantitative imaging features and 3 hemodynamic parameters [mean pulmonary artery pressure, right atrial pressure (RAP), pulmonary artery oxygen saturation (PA SaO2)] were gathered. Diameter-based CT features were measured in axial and adjusted multiplane reconstructions (MPR). Univariate analysis was performed for qualitative and quantitative features. A random forest algorithm was trained on imaging features to predict hemodynamic risk groups. Feature importance was calculated for all models. Qualitative and quantitative parameters showed no significant differences between ECG and non-ECG gated CTs. Depending on reconstruction plane, five quantitative features were significantly different, but mean absolute difference between parameters (MPR vs. axial) was 0.3 mm with no difference in correlation with hemodynamic parameters. Univariate analysis showed moderate to strong correlation for multiple imaging features with hemodynamic parameters. The model achieved an AUC score of 0.82 for the mPAP based risk stratification and 0.74 for the PA SaO2 risk stratification. Contrast agent retention in hepatic vein, mosaic attenuation pattern and the ratio right atrium/left ventricle were the most important features among other parameters. Quantitative and qualitative imaging features of reconstructions correlate with hemodynamic parameters in preoperative CTEPH patients-regardless of MPR adaption. Machine learning based analysis of preoperative imaging features can be used for non-invasive risk stratification. Qualitative features seem to be more important than previously anticipated.

Development and Validation of a Deep Learning Algorithm to Differentiate Colon Carcinoma From Acute Diverticulitis in Computed Tomography Images.

Importance: Differentiating between malignant and benign etiology in large-bowel wall thickening on computed tomography (CT) images can be a challenging task. Artificial intelligence (AI) support systems can improve the diagnostic accuracy of radiologists, as shown for a variety of imaging tasks. Improvements in diagnostic performance, in particular the reduction of false-negative findings, may be useful in patient care. Objective: To develop and evaluate a deep learning algorithm able to differentiate colon carcinoma (CC) and acute diverticulitis (AD) on CT images and analyze the impact of the AI-support system in a reader study. Design, Setting, and Participants: In this diagnostic study, patients who underwent surgery between July 1, 2005, and October 1, 2020, for CC or AD were included. Three-dimensional (3-D) bounding boxes including the diseased bowel segment and surrounding mesentery were manually delineated and used to develop a 3-D convolutional neural network (CNN). A reader study with 10 observers of different experience levels was conducted. Readers were asked to classify the testing cohort under reading room conditions, first without and then with algorithmic support. Main Outcomes and Measures: To evaluate the diagnostic performance, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for all readers and reader groups with and without AI support. Metrics were compared using the McNemar test and relative and absolute predictive value comparisons. Results: A total of 585 patients (AD: n = 267, CC: n = 318; mean [SD] age, 63.2 [13.4] years; 341 men [58.3%]) were included. The 3-D CNN reached a sensitivity of 83.3% (95% CI, 70.0%-96.6%) and specificity of 86.6% (95% CI, 74.5%-98.8%) for the test set, compared with the mean reader sensitivity of 77.6% (95% CI, 72.9%-82.3%) and specificity of 81.6% (95% CI, 77.2%-86.1%). The combined group of readers improved significantly with AI support from a sensitivity of 77.6% to 85.6% (95% CI, 81.3%-89.3%; P < .001) and a specificity of 81.6% to 91.3% (95% CI, 88.1%-94.5%; P < .001). Artificial intelligence support significantly reduced the number of false-negative and false-positive findings (NPV from 78.5% to 86.4% and PPV from 80.9% to 90.8%; P < .001). Conclusions and Relevance: The findings of this study suggest that a deep learning model able to distinguish CC and AD in CT images as a support system may significantly improve the diagnostic performance of radiologists, which may improve patient care.

CT Attenuation of Hepatic Pancreatic Cancer Metastases Correlates with Prognostically Detrimental Metastatic Necrosis.

Percutaneous CT-guided biopsy is a frequently performed procedure for the confirmation and molecular workup of hepatic metastases of pancreatic ductal adenocarcinoma (PDAC). Tumor necrosis of primary PDAC has shown a negative prognostic impact in recent studies. This study aims to examine predictability in CT scans and the prognostic impact of necrosis in hepatic metastases of PDAC. In this tertiary-center retrospective cohort study, we included 36 patients with hepatic metastases of PDAC who underwent CT-guided hepatic biopsies. Normalized attenuation of the biopsied metastasis was determined in venous phase contrast-enhanced planning scans obtained prior to biopsy by automatic, threshold-based 3D segmentation and manual, blinded 2D segmentation. A board-certified pathologist specialized in hepatic pathology histologically quantified the tumor necrosis and cellularity of the biopsy cylinders. We found a significant inverse-linear correlation between normalized attenuation and the fraction of necrosis (Pearson's r = 0.51, p < 0.001 for automatic 3D segmentation or Pearson's r = 0.52, p < 0.001 for manual 2D segmentation), whereas no correlation was found with tumor cellularity. Additionally, we discovered that patients with a fraction of necrosis ≥ 20% in metastases had a significantly shorter overall survival (p < 0.035). In summary, tumor necrosis of PDAC metastases can be estimated from contrast-enhanced CT scans, which could help to improve biopsy sample pattern planning. In addition, liver metastatic necrosis may serve as a prognostic biomarker in PDAC.

Cost-Effectiveness of Artificial Intelligence Support in Computed Tomography-Based Lung Cancer Screening.

BACKGROUND: Lung cancer screening is already implemented in the USA and strongly recommended by European Radiological and Thoracic societies as well. Upon implementation, the total number of thoracic computed tomographies (CT) is likely to rise significantly. As shown in previous studies, modern artificial intelligence-based algorithms are on-par or even exceed radiologist's performance in lung nodule detection and classification. Therefore, the aim of this study was to evaluate the cost-effectiveness of an AI-based system in the context of baseline lung cancer screening. METHODS: In this retrospective study, a decision model based on Markov simulation was developed to estimate the quality-adjusted life-years (QALYs) and lifetime costs of the diagnostic modalities. Literature research was performed to determine model input parameters. Model uncertainty and possible costs of the AI-system were assessed using deterministic and probabilistic sensitivity analysis. RESULTS: In the base case scenario CT + AI resulted in a negative incremental cost-effectiveness ratio (ICER) as compared to CT only, showing lower costs and higher effectiveness. Threshold analysis showed that the ICER remained negative up to a threshold of USD 68 for the AI support. The willingness-to-pay of USD 100,000 was crossed at a value of USD 1240. Deterministic and probabilistic sensitivity analysis showed model robustness for varying input parameters. CONCLUSION: Based on our results, the use of an AI-based system in the initial low-dose CT scan of lung cancer screening is a feasible diagnostic strategy from a cost-effectiveness perspective.

Feature Robustness and Diagnostic Capabilities of Convolutional Neural Networks Against Radiomics Features in Computed Tomography Imaging.

MATERIALS AND METHODS: Imaging phantoms were scanned twice on 3 computed tomography scanners from 2 different manufactures with varying tube voltages and currents. Phantoms were segmented, and features were extracted using PyRadiomics and a pretrained CNN. After standardization the concordance correlation coefficient (CCC), mean feature variance, feature range, and the coefficient of variant were calculated to assess feature robustness. In addition, the cosine similarity was calculated for the vectorized activation maps for an exemplary phantom. For the in vivo comparison, the radiomics and CNN features of 30 patients with hepatocellular carcinoma (HCC) and 30 patients with hepatic colon carcinoma metastasis were compared. RESULTS: In total, 851 radiomics features and 256 CNN features were extracted for each phantom. For all phantoms, the global CCC of the CNN features was above 98%, whereas the highest CCC for the radiomics features was 36%. The mean feature variance and feature range was significantly lower for the CNN features. Using a coefficient of variant ≤0.2 as a threshold to define robust features and averaging across all phantoms 346 of 851 (41%) radiomics features and 196 of 256 (77%) CNN features were found to be robust. The cosine similarity was greater than 0.98 for all scanner and parameter variations. In the retrospective analysis, 122 of the 256 CNN (49%) features showed significant differences between HCC and hepatic colon metastasis. DISCUSSION: Convolutional neural network features were more stable compared with radiomics features against technical variations. Moreover, the possibility of tumor entity differentiation based on CNN features was shown. Combined with visualization methods, CNN features are expected to increase reproducibility of quantitative image representations. Further studies are warranted to investigate the impact of feature stability on radiological image-based prediction of clinical outcomes.

High-Resolution, High b-Value Computed Diffusion-Weighted Imaging Improves Detection of Pancreatic Ductal Adenocarcinoma.

BACKGROUND: Our purpose was to investigate the potential of high-resolution, high b-value computed DWI (cDWI) in pancreatic ductal adenocarcinoma (PDAC) detection. MATERIALS AND METHODS: We retrospectively enrolled 44 patients with confirmed PDAC. Respiratory-triggered, diffusion-weighted, single-shot echo-planar imaging (ss-EPI) with both conventional (i.e., full field-of-view, 3 × 3 × 4 mm voxel size, b = 0, 50, 300, 600 s/mm2) and high-resolution (i.e., reduced field-of-view, 2.5 × 2.5 × 3 mm voxel size, b = 0, 50, 300, 600, 1000 s/mm2) imaging was performed for suspected PDAC. cDWI datasets at b = 1000 s/mm2 were generated for the conventional and high-resolution datasets. Three radiologists were asked to subjectively rate (on a Likert scale of 1-4) the following metrics: image quality, lesion detection and delineation, and lesion-to-pancreas intensity relation. Furthermore, the following quantitative image parameters were assessed: apparent signal-to-noise ratio (aSNR), contrast-to-noise ratio (aCNR), and lesion-to-pancreas contrast ratio (CR). RESULTS: High-resolution, high b-value computed DWI (r-cDWI1000) enabled significant improvement in lesion detection and a higher incidence of a high lesion-to-pancreas intensity relation (type 1, clear hyperintense) compared to conventional high b-value computed and high-resolution high b-value acquired DWI (f-cDWI1000 and r-aDWI1000, respectively). Image quality was rated inferior in the r-cDWI1000 datasets compared to r-aDWI1000. Furthermore, the aCNR and CR were higher in the r-cDWI1000 datasets than in f-cDWI1000 and r-aDWI1000. CONCLUSION: High-resolution, high b-value computed DWI provides significantly better visualization of PDAC compared to the conventional high b-value computed and high-resolution high b-value images acquired by DWI.

Prospectively Accelerated T2-Weighted Imaging of the Prostate by Combining Compressed SENSE and Deep Learning in Patients with Histologically Proven Prostate Cancer.

BACKGROUND: To assess the performance of prospectively accelerated and deep learning (DL) reconstructed T2-weighted (T2w) imaging in volunteers and patients with histologically proven prostate cancer (PCa). METHODS: Prospectively undersampled T2w datasets were acquired with acceleration factors of 1.7 (reference), 3.4 and 4.8 in 10 healthy volunteers and 23 patients with histologically proven PCa. Image reconstructions using compressed SENSE (C-SENSE) and a combination of C-SENSE and DL-based artificial intelligence (C-SENSE AI) were analyzed. Qualitative image comparison was performed using a 6-point Likert scale (overall image quality, noise, motion artifacts, lesion detection, diagnostic certainty); the T2 and PI-RADS scores were compared between the two reconstructions. Additionally, quantitative image parameters were assessed (apparent SNR, apparent CNR, lesion size, line profiles). RESULTS: All C-SENSE AI-reconstructed images received a significantly higher qualitative rating compared to the C-SENSE standard images. Analysis of the quantitative parameters supported this finding, with significantly higher aSNR and aCNR. The line profiles demonstrated a significantly steeper signal change at the border of the prostatic lesion and the adjacent normal tissue in the C-SENSE AI-reconstructed images, whereas the T2 and PI-RADS scores as well as the lesion size did not differ. CONCLUSION: In this prospective study, we demonstrated the clinical feasibility of a novel C-SENSE AI reconstruction enabling a 58% acceleration in T2w imaging of the prostate while obtaining significantly better image quality.

Beyond the d-dimer - Machine-learning assisted pre-test probability evaluation in patients with suspected pulmonary embolism and elevated d-dimers.

INTRODUCTION: Acute pulmonary embolism (PE) is a leading cardiovascular cause of death, resembling a common indication for emergency computed tomography (CT). Nonetheless, in clinical routine most CTs performed for suspicion of PE excluded the suspected diagnosis. As patients with low to intermediate risk for PE are triaged according to the d-dimer, its relatively low specifity and widespread elevation among elderly might be an underlying issue. Aim of this study was to find potential predictors based on initial emergency blood tests in patients with elevated d-dimers and suspected PE to further increase pre-test probability. METHODS: In this retrospective study all patients at the local university hospital's emergency room from 2009 to 2019 with suspected PE, emergency blood testing and CT were included. Cluster analysis was performed to separate groups with distinct laboratory parameter profiles and PE frequencies were compared. Machine learning algorithms were trained on the groups to predict individual PE probability based on emergency laboratory parameters. RESULTS: Overall, PE frequency among the 2045 analyzed patients was 41%. Three clusters with significant differences (p ≤ 0.05) in PE frequency were identified: C1 showed a PE frequency of 43%, C2 40% and C3 33%. Laboratory parameter profiles (e.g. creatinine) differed significantly between clusters (p ≤ 0.0001). Both logistic regression and support-vector machines were able to predict clusters with an accuracy of over 90%. DISCUSSION: Initial blood parameters seem to enable further differentiation of patients with suspected PE and elevated d-dimers to raise pre-test probability of PE. Machine-learning-based prediction models might help to further narrow down CT indications in the future.

Efficient, high-performance semantic segmentation using multi-scale feature extraction.

The success of deep learning in recent years has arguably been driven by the availability of large datasets for training powerful predictive algorithms. In medical applications however, the sensitive nature of the data limits the collection and exchange of large-scale datasets. Privacy-preserving and collaborative learning systems can enable the successful application of machine learning in medicine. However, collaborative protocols such as federated learning require the frequent transfer of parameter updates over a network. To enable the deployment of such protocols to a wide range of systems with varying computational performance, efficient deep learning architectures for resource-constrained environments are required. Here we present MoNet, a small, highly optimized neural-network-based segmentation algorithm leveraging efficient multi-scale image features. MoNet is a shallow, U-Net-like architecture based on repeated, dilated convolutions with decreasing dilation rates. We apply and test our architecture on the challenging clinical tasks of pancreatic segmentation in computed tomography (CT) images as well as brain tumor segmentation in magnetic resonance imaging (MRI) data. We assess our model's segmentation performance and demonstrate that it provides performance on par with compared architectures while providing superior out-of-sample generalization performance, outperforming larger architectures on an independent validation set, while utilizing significantly fewer parameters. We furthermore confirm the suitability of our architecture for federated learning applications by demonstrating a substantial reduction in serialized model storage requirement as a surrogate for network data transfer. Finally, we evaluate MoNet's inference latency on the central processing unit (CPU) to determine its utility in environments without access to graphics processing units. Our implementation is publicly available as free and open-source software.

Prediction of Tumor Cellularity in Resectable PDAC from Preoperative Computed Tomography Imaging.

BACKGROUND: PDAC remains a tumor entity with poor prognosis and a 5-year survival rate below 10%. Recent research has revealed invasive biomarkers, such as distinct molecular subtypes, predictive for therapy response and patient survival. Non-invasive prediction of individual patient outcome however remains an unresolved task. METHODS: Discrete cellularity regions of PDAC resection specimen (n = 43) were analyzed by routine histopathological work up. Regional tumor cellularity and CT-derived Hounsfield Units (HU, n = 66) as well as iodine concentrations were regionally matched. One-way ANOVA and pairwise t-tests were performed to assess the relationship between different cellularity level in conventional, virtual monoenergetic 40 keV (monoE 40 keV) and iodine map reconstructions. RESULTS: A statistically significant negative correlation between regional tumor cellularity in histopathology and CT-derived HU from corresponding image regions was identified. Radiological differentiation was best possible in monoE 40 keV CT images. However, HU values differed significantly in conventional reconstructions as well, indicating the possibility of a broad clinical application of this finding. CONCLUSION: In this study we establish a novel method for CT-based prediction of tumor cellularity for in-vivo tumor characterization in PDAC patients.

Lung nodule detection in chest X-rays using synthetic ground-truth data comparing CNN-based diagnosis to human performance.

We present a method to generate synthetic thorax radiographs with realistic nodules from CT scans, and a perfect ground truth knowledge. We evaluated the detection performance of nine radiologists and two convolutional neural networks in a reader study. Nodules were artificially inserted into the lung of a CT volume and synthetic radiographs were obtained by forward-projecting the volume. Hence, our framework allowed for a detailed evaluation of CAD systems' and radiologists' performance due to the availability of accurate ground-truth labels for nodules from synthetic data. Radiographs for network training (U-Net and RetinaNet) were generated from 855 CT scans of a public dataset. For the reader study, 201 radiographs were generated from 21 nodule-free CT scans with altering nodule positions, sizes and nodule counts of inserted nodules. Average true positive detections by nine radiologists were 248.8 nodules, 51.7 false positive predicted nodules and 121.2 false negative predicted nodules. The best performing CAD system achieved 268 true positives, 66 false positives and 102 false negatives. Corresponding weighted alternative free response operating characteristic figure-of-merits (wAFROC FOM) for the radiologists range from 0.54 to 0.87 compared to a value of 0.81 (CI 0.75-0.87) for the best performing CNN. The CNN did not perform significantly better against the combined average of the 9 readers (p = 0.49). Paramediastinal nodules accounted for most false positive and false negative detections by readers, which can be explained by the presence of more tissue in this area.