BSBerend C. Stoel

Berend C. Stoel

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Berend C. Stoel has a total of 140 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Aggregation of Class Activation Maps for Explaining Deep Learning at a Population Level

    by Yanli Li, Xikai Tang, Dennis Ton, Denis P. Shamonin, Hessam Sokooti, Monique Reijnierse, Annette H.M. van der Helm-Van Mil, Hans Reiber, Berend C. Stoel

    Background and Objective. Class activation maps (CAMs) are frequently used to help understand the inference by deep learning (DL) models, especially in medical imaging that requires high reliability and explainability. Because a single CAM only provides qualitative insight from one case, population-level validation requires aggregation of CAMs. Currently, this aggregation is performed manually in observer studies, which are time-consuming and potentially biased.Methods. We propose an automatic CAM aggregation workflow, based on image segmentation. Objective validation of the aggregation method requires ground truth regarding the importance of specific image patterns for classification. We define this importance as the posterior probability of a class given the presence of specific pathological regions. A series of simulated classification tasks was designed with predefined posterior probabilities for virtual image patterns. We evaluated whether mean CAM activation within pathological regions consistently reflected these probabilities.Results. A high reliability and sensitivity was demonstrated, showing a strong correlation (r = 0.98) between known posterior probabilities and regional mean activations. The results also demonstrate that the CAM aggregation can generate hypotheses about the causality between these regions and the clinical outcomes. The validated aggregation method was then applied to a rheumatoid arthritis-related dataset, to prove its potential applications in clinical practice.Conclusions. We expect that the proposed CAM aggregation can help validate DL models at a population level and reveal possible correlations between regions containing specific image patterns and clinical outcomes.

  • Open Access

    Localization of Early Treatment Effects From Magnetic Resonance Imaging of the Wrist, Hand, and Foot of Patients With Arthralgia at Risk for Rheumatoid Arthritis

    by Jingnan Jia, Tahereh Hassanzadeh, Yanli Li, Denis P. Shamonin, Dennis A. Ton, M. Els Bakker, Monique Reijnierse, Annette H. M. van der Helm-Van Mil, Berend C. Stoel

  • Open Access

    Predictive value of aorta enhancement on computed tomographic pulmonary angiography in pulmonary embolism

    by Qiuyu Du, Sophie N. M. ter Haar, Jingnan Jia, Lucia J. M. Kroft, Marius Staring, Frederikus A. Klok, Berend C. Stoel

    Background: Pulmonary embolism (PE) is a life-threatening condition requiring prompt diagnosis and treatment. Visual assessment of computed tomographic pulmonary angiography (CTPA) is the first-choice diagnostic tool. New imaging biomarkers could provide additional prognostic information for improved risk stratification. We hypothesized in this exploratory study, that contrast enhancement patterns in the aorta may contain such information. Methods: CTPA scans of 93 acute PE patients were analyzed retrospectively. Firstly, the aorta was segmented automatically by TotalSegmentator and its centerline was extracted. Subsequently, lines were fitted on intensities within a region of interest perpendicularly to the aorta centerline, from which three parameters were extracted: mean intensity, proximal intensity and contrast gradient. After confounder analysis, logistic regression with forward selection evaluated the predictive value of these parameters for 12 adverse outcomes (six short-term and six long-term). Results: Lung volume, aorta dimension and contrast delay were considered as possible confounders but were not selected by forward selection. Logistic regression (n = 93) showed that a less steep contrast gradient (decreasing by 10 Hounsfield unit/%) was associated with a reduction in odds of the following short-term adverse outcomes: 48.1% for intensive care unit admission (odds ratio [OR] = 0.519, 95% confidence interval [CI]: 0.306–0.804), 29.3% for oxygen therapy >24 hours (OR = 0.707, 95% CI: 0.496–0.976), 60.6% for reperfusion therapy (OR = 0.394, 95% CI: 0.178–0.682), 57.5% for vasopressor therapy (OR = 0.425, 95% CI: 0.194–0.741), and 50.2% for PE-related death (OR = 0.498, 95% CI: 0.246–0.915). No significant associations were found with long-term outcomes. Conclusions: The aorta contrast gradient, automatically quantified from CTPA, is a relevant adjunctive predictor for short-term outcomes in PE patients. Long-term outcomes, however, could not be predicted by aorta measurement. This pilot study provides initial insights into predictive value of aorta enhancement, stimulating further exploration with external data.

  • Open Access

    Automatic joint inflammation estimation based on regression neural networks

    by Yanli Li, Dennis A. Ton, Denis P. Shamonin, Monique Reijnierse, Annette H. M. van der Helm‐Van Mil, Berend C. Stoel

    Background: Quantitative assessment of inflammation from hand and forefoot MRI scans is crucial for evaluating the severity, progression, and treatment response in inflammatory disease like rheumatoid arthritis (RA). Traditionally, this relies on visual evaluation of signs like bone marrow edema (BME), tenosynovitis, and synovitis, which is time‐consuming, subjective, and prone to inherent inter/intra‐reader variability. Purpose: This study aims at an automatic DL‐based MRI analysis of inflammatory signs in RA system for inflammation assessment to facilitate related diagnoses and studies. Methods: We developed an A utomatic D L‐based M RI analysis of I nflammatory signs in RA (ADMIRA) system for inflammation assessment, using pre‐ and post‐processing alongside DL models to estimate inflammation scores from fat saturated, contrast‐enhanced T1‐weighted MRI scans of 2254 subjects across four study populations. These MRI scans include three different anatomical sites, wrist, metacarpophalangeal (MCP) and metatarsophalangeal (MTP) joints, as the objects for inflammation assessment. The scans were divided into training, monitoring, testing and validation sets to ensure robust performance evaluation, using Pearson's correlation coefficients and Intra‐class correlation coefficients. A revised class activation mapping (CAM) algorithm was used to validate the DL model's reliability, illustrating its inference process. Results: The system achieved mean R/ICCs of nearly 0.9 for synovitis and tenosynovitis on test sets and 0.8 on the validation set, with slightly lower scores for BME (0.8 and 0.7, respectively). This system presents a performance close to human experts on the same datasets. Meanwhile, the visualization results indicate the DL models have a inference process consistent with expert knowledge. Conclusions: Results show that ADMIRA provides accurate, expert‐level inflammation estimation, particularly for synovitis and tenosynovitis, offering a fast, reliable alternative to manual methods for RA monitoring and analysis. We expect that this automatic method could help to reduce labor costs and improve the efficiency of diagnosis in the future.


  • Open Access

    Feature analysis for proper intensity scaling and feature distinction in class activation maps

    by Yanli Li, Denis P. Shamonin, Tahereh Hassanzadeh, Monique Reijnierse, Annette H.M. van der Helm-Van Mil, Berend C. Stoel


  • Open Access

    Explainable fully automated CT scoring of interstitial lung disease for patients suspected of systemic sclerosis by cascaded regression neural networks and its comparison with experts

    by Jingnan Jia, Irene Hernández-Girón, Anne A. Schouffoer, Jeska K. de Vries-Bouwstra, Maarten K. Ninaber, Julie C. Korving, Marius Staring, Lucia J. M. Kroft, Berend C. Stoel

    Visual scoring of interstitial lung disease in systemic sclerosis (SSc-ILD) from CT scans is laborious, subjective and time-consuming. This study aims to develop a deep learning framework to automate SSc-ILD scoring. The automated framework is a cascade of two neural networks. The first network selects the craniocaudal positions of the five scoring levels. Subsequently, for each level, the second network estimates the ratio of three patterns to the total lung area: the total extent of disease (TOT), ground glass (GG) and reticulation (RET). To overcome the score imbalance in the second network, we propose a method to augment the training dataset with synthetic data. To explain the network’s output, a heat map method is introduced to highlight the candidate interstitial lung disease regions. The explainability of heat maps was evaluated by two human experts and a quantitative method that uses the heat map to produce the score. The results show that our framework achieved a \(\kappa\) of 0.66, 0.58, and 0.65, for the TOT, GG and RET scoring, respectively. Both experts agreed with the heat maps in 91%, 90% and 80% of cases, respectively. Therefore, it is feasible to develop a framework for automated SSc-ILD scoring, which performs competitively with human experts and provides high-quality explanations using heat maps. Confirming the model’s generalizability is needed in future studies.


  • Open Access

    Accuracy and precision of Volumetric Matching Micromotion Analysis (V3MA) is similar to RSA for tibial component migration in TKA

    by Nienke N. de Laat, Lennard A. Koster, Berend C. Stoel, Rob G. H. H. Nelissen, Bart L. Kaptein

    Radiostereometric analysis (RSA) is the current gold standard to determine implant migration, but it requires bone markers and special equipment. Therefore, we developed VoluMetric Matching Micromotion Analysis (V3MA), a software program for Computed Tomography‐based radiostereometric analysis (CT‐RSA). This study aimed to determine the accuracy and precision of V3MA in vitro compared to RSA and provide a clinical proof of concept. The accuracy (RMSE (Root Mean Squared Error)) and precision (SD (standard deviation)) of V3MA were compared to RSA. A tibial component was placed in 21 different positions within a cadaveric bone to assess accuracy. For precision, a total of 20 repeated zero‐migration examinations from 4 cadaveric bones with cemented tibial components were performed. In 6 total knee arthroplasty (TKA) patients 1 to 5 year migration was measured with V3MA and RSA. V3MA accuracy ranged between 0.02 and 0.09 mm for translations and was 0.01° for internal–external rotations. For RSA, the accuracy ranged between 0.03 and 0.09 mm for translations and was 0.09° for internal‐external rotations. V3MA precision ranged between 0.01 and 0.06 mm for translations and 0.02 to 0.07° for rotations. RSA precision ranged between 0.00 and 0.06 mm for translations and 0.04 to 0.25° for rotations. V3MA was successful in 6 clinical cases and no systematic bias was present. In conclusion, the accuracy and precision of V3MA were similar to RSA. Therefore, V3MA is a promising alternative to RSA in migration measurements of tibial components in TKA.


  • Open Access

    Using 3D point cloud and graph-based neural networks to improve the estimation of pulmonary function tests from chest CT

    by Jingnan Jia, Bo Yu, Prerak Mody, Maarten K. Ninaber, Anne A. Schouffoer, Jeska K. de Vries-Bouwstra, Lucia J.M. Kroft, Marius Staring, Berend C. Stoel


  • Open Access

    seg-metrics: a Python package to compute segmentation metrics

    by Jingnan Jia, Marius Staring, Berend C. Stoel

    Medical image segmentation (MIS) is an important task in medical image processing. Unfortunately, there is not a out-of-the-box python package for the evaluation metrics of MIS. Therefore, we developed seg-metrics , an open-source Python package for MIS model evaluation. Unlike existing packages, seg-metrics offers user-friendly interfaces for various overlap-based and distance-based metrics, providing a comprehensive solution. seg-metrics supports multiple file formats and is easily installable through the Python Package Index (PyPI). With a focus on speed and convenience, seg-metrics stands as a valuable tool for efficient MIS model assessment.


  • Deep learning in rheumatological image interpretation

    by Berend C. Stoel, Marius Staring, Monique Reijnierse, Annette H. M. van der Helm-Van Mil

    Artificial intelligence techniques, specifically deep learning, have already affected daily life in a wide range of areas. Likewise, initial applications have been explored in rheumatology. Deep learning might not easily surpass the accuracy of classic techniques when performing classification or regression on low-dimensional numerical data. With images as input, however, deep learning has become so successful that it has already outperformed the majority of conventional image-processing techniques developed during the past 50 years. As with any new imaging technology, rheumatologists and radiologists need to consider adapting their arsenal of diagnostic, prognostic and monitoring tools, and even their clinical role and collaborations. This adaptation requires a basic understanding of the technical background of deep learning, to efficiently utilize its benefits but also to recognize its drawbacks and pitfalls, as blindly relying on deep learning might be at odds with its capabilities. To facilitate such an understanding, it is necessary to provide an overview of deep-learning techniques for automatic image analysis in detecting, quantifying, predicting and monitoring rheumatic diseases, and of currently published deep-learning applications in radiological imaging for rheumatology, with critical assessment of possible limitations, errors and confounders, and conceivable consequences for rheumatologists and radiologists in clinical practice.