Labor Regensburg Medical Image Computing (ReMIC)
Refine
Year of publication
Document Type
- Article (70)
- conference proceeding (article) (58)
- conference proceeding (presentation, abstract) (40)
- conference proceeding (volume) (9)
- Preprint (7)
- Research Data (2)
- Book (1)
- Part of a Book (1)
- Doctoral Thesis (1)
- Other (1)
Is part of the Bibliography
- no (190)
Keywords
- Bildgebendes Verfahren (23)
- Deep Learning (21)
- Diagnose (15)
- Künstliche Intelligenz (14)
- Artificial Intelligence (13)
- Maschinelles Lernen (12)
- Bildverarbeitung (11)
- Gehirn (11)
- Computerunterstützte Medizin (10)
- Kernspintomografie (9)
Institute
- Labor Regensburg Medical Image Computing (ReMIC) (190)
- Fakultät Informatik und Mathematik (189)
- Regensburg Center of Biomedical Engineering - RCBE (48)
- Regensburg Center of Health Sciences and Technology - RCHST (45)
- Fakultät Sozial- und Gesundheitswissenschaften (3)
- Labor Biomechanik (LBM) (2)
- Labor Empirische Sozialforschung (2)
- Labor für Technikfolgenabschätzung und Angewandte Ethik (LaTe) (2)
- Fakultät Elektro- und Informationstechnik (1)
- Fakultät Maschinenbau (1)
Begutachtungsstatus
- peer-reviewed (108)
- begutachtet (2)
Note: A script for extracting the individual frames from the video files while preserving the challenge-compliant directory structure and frame-to-mask naming conventions is available on GitHub and can be accessed here: https://github.com/remic-othr/PhaKIR_Dataset.
The dataset is described in the following publications:
Rueckert, Tobias et al.: Comparative validation of surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation in endoscopy: Results of the PhaKIR 2024 challenge. arXiv preprint, https://arxiv.org/abs/2507.16559. 2025.
Rueckert, Tobias et al.: Video Dataset for Surgical Phase, Keypoint, and Instrument Recognition in Laparoscopic Surgery (PhaKIR). arXiv preprint, https://arxiv.org/abs/2511.06549. 2025.
The proposed dataset was used as the training dataset in the PhaKIR challenge (https://phakir.re-mic.de/) as part of EndoVis-2024 at MICCAI 2024 and consists of eight real-world videos of human cholecystectomies ranging from 23 to 60 minutes in duration. The procedures were performed by experienced physicians, and the videos were recorded in three hospitals. In addition to existing datasets, our annotations provide pixel-wise instance segmentation masks of surgical instruments for a total of 19 categories, coordinates of relevant instrument keypoints (instrument tip(s), shaft-tip transition, shaft), both at an interval of one frame per second, and specifications regarding the intervention phases for a total of eight different phase categories for each individual frame in one dataset and thus comprehensively cover instrument localization and the context of the operation. Furthermore, the provision of the complete video sequences offers the opportunity to include the temporal information regarding the respective tasks and thus further optimize the resulting methods and outcomes.
This dataset is a subset of the original CATARACTS test dataset and is used by the OpenMIBOOD framework to evaluate a specific out-of-distribution setting.
When using this dataset, it is mandatory to cite the corresponding publication (OpenMIBOOD (10.1109/CVPR52734.2025.02410)) and follow the acknowledgement and citation requirements of the original dataset (CATARACTS).
The original CATARACTS dataset (associated publication,Homepage) consists of 50 videos of cataract surgeries, split into 25 train and 25 test videos.
This subset contains the frames of the first 5 test videos. Further, black frames at the beginning of each video were removed.
Reliable recognition and localization of surgical instruments in endoscopic video recordings are foundational for a wide range of applications in computer- and robot-assisted minimally invasive surgery (RAMIS), including surgical training, skill assessment, and autonomous assistance. However, robust performance under real-world conditions remains a significant challenge. Incorporating surgical context – such as the current procedural phase – has emerged as a promising strategy to improve robustness and interpretability.
To address these challenges, we organized the Surgical Procedure Phase, Keypoint, and Instrument Recognition (PhaKIR) sub-challenge as part of the Endoscopic Vision (EndoVis) challenge at MICCAI 2024. We introduced a novel, multi-center dataset comprising thirteen full-length laparoscopic cholecystectomy videos collected from three distinct medical institutions, with unified annotations for three interrelated tasks: surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation. Unlike existing datasets, ours enables joint investigation of instrument localization and procedural context within the same data while supporting the integration of temporal information across entire procedures.
We report results and findings in accordance with the BIAS guidelines for biomedical image analysis challenges. The PhaKIR sub-challenge advances the field by providing a unique benchmark for developing temporally aware, context-driven methods in RAMIS and offers a high-quality resource to support future research in surgical scene understanding.
DIY challenge blueprint: from organization to technical realization in biomedical image analysis
(2025)
Biomedical image analysis challenges have become the de facto standard for publishing new datasets and benchmarking different state-of-the-art algorithms. Most challenges use commercial cloud-based platforms, which can limit custom options and involve disadvantages such as reduced data control and increased costs for extended functionalities. In contrast, Do-It-Yourself (DIY) approaches have the capability to emphasize reliability, compliance, and custom features, providing a solid basis for low-cost, custom designs in self-hosted systems. Our approach emphasizes cost efficiency, improved data sovereignty, and strong compliance with regulatory frameworks, such as the GDPR. This paper presents a blueprint for DIY biomedical imaging challenges, designed to provide institutions with greater autonomy over their challenge infrastructure. Our approach comprehensively addresses both organizational and technical dimensions, including key user roles, data management strategies, and secure, efficient workflows. Key technical contributions include a modular, containerized infrastructure based on Docker, integration of open-source identity management, and automated solution evaluation workflows. Practical deployment guidelines are provided to facilitate implementation and operational stability. The feasibility and adaptability of the proposed framework are demonstrated through the MICCAI 2024 PhaKIR challenge with multiple international teams submitting and validating their solutions through our self-hosted platform. This work can be used as a baseline for future self-hosted DIY implementations and our results encourage further studies in the area of biomedical image analysis challenges.
Purpose
Recognizing previously unseen classes with neural networks is a significant challenge due to their limited generalization capabilities. This issue is particularly critical in safety-critical domains such as medical applications, where accurate classification is essential for reliability and patient safety. Zero-shot learning methods address this challenge by utilizing additional semantic data, with their performance relying heavily on the quality of the generated embeddings.
Methods
This work investigates the use of full descriptive sentences, generated by a Sentence-BERT model, as class representations, compared to simpler category-based word embeddings derived from a BERT model. Additionally, the impact of z-score normalization as a post-processing step on these embeddings is explored. The proposed approach is evaluated on a multi-label generalized zero-shot learning task, focusing on the recognition of surgical instruments in endoscopic images from minimally invasive cholecystectomies.
Results
The results demonstrate that combining sentence embeddings and z-score normalization significantly improves model performance. For unseen classes, the AUROC improves from 43.9% to 64.9%, and the multi-label accuracy from 26.1% to 79.5%. Overall performance measured across both seen and unseen classes improves from 49.3% to 64.9% in AUROC and from 37.3% to 65.1% in multi-label accuracy, highlighting the effectiveness of our approach.
Conclusion
These findings demonstrate that sentence embeddings and z-score normalization can substantially enhance the generalization performance of zero-shot learning models. However, as the study is based on a single dataset, future work should validate the method across diverse datasets and application domains to establish its robustness and broader applicability.
Einleitung: Die endoskopische Submukosadissektion (ESD) ist ein komplexes endoskopisches Verfahren, das technische Expertise erfordert. Objektive Methoden zur Analyse von interventionellen Abläufen bei ESD könnten für Qualitätssicherung und Ausbildung, wie auch eine automatische Befunderstellung von Nutzen sein.
Ziele: In dieser Studie wurde ein KI-Algorithmus für die Erkennung und Klassifizierung der interventionellen Phasen der ESD entwickelt, um die technische Basis für eine standardisierte Leistungsbewertung und automatische Befunderstellung zu schaffen.
Methodik: Vollständige ESD-Videoaufnahmen von 49 Patienten wurden retrospektiv zusammengestellt. Der Datensatz umfasste 6.390.151 Einzelbilder, die alle für die folgenden interventionellen Phasen annotiert wurden: Diagnostik, Markierung, Injektion, Dissektion und Hämostase. 3.973.712 Bilder (28 Patienten) wurden für das Training eines Video-Swin-Transformers genutzt. Dabei wurde temporale Information durch standardisierte BIldextraktion in festgelegten zeitlichen Abständen zum analysierten Bild inkorporiert. 2.416.439 separate Bilder (21 Patienten) wurden für eine interne Validierung genutzt.
Ergebnis: Bei der internen Evaluation erreichte das System insgesamt einen F1-Wert von 0,88. Es wurden F1-Werte von 0,99, 0,89, 0,89, 0,91 und 0,52 für Diagnostik, Markierung, Injektion, Dissektion bzw. Blutungsmanagement gemessen. Die Sensitivitäten für dieselben Parameter betrugen 1,00, 0,80, 0,94, 0,89 und 0,67, die Spezifitäten lagen bei 1,00, 1,00, 0,98, 0,88 und 0,93. Positive prädiktive Werte wurden mit 0,98, 1,00, 0,85, 0,94 und 0,43 gemessen.
Schlussfolgerung: In dieser vorläufigen Studie zeigte ein KI-Algorithmus eine hohe Leistungsfähigkeit für die Einzelbild-Erkennung von Verfahrensphasen während der ESD. Die vergleichsweise niedrige Leistung für die Blutungsphase wurde auf das seltene Auftreten von Blutungsepisoden im Trainingsdatensatz zurückgeführt, der zu diesem Zeitpunkt nur Videos in voller Länge umfasste. Die zukünftige Entwicklung des Algorithmus wird sich auf die Reduzierung von Klassenungleichgewichten durch selektive Annotationsprotokolle konzentrieren.
Instrumentenerkennung während der endoskopischen Submukosadissektion mittels künstlicher Intelligenz
(2025)
Einleitung: Die endoskopische Submukosadissektion (ESD) ist eine komplexe Technik zur Resektion gastrointestinaler Frühneoplasien. Dabei werden für die verschiedenen Schritte der Intervention spezifische endoskopische Instrumente verwendet. Die präzise und automatische Erkennung und Abgrenzung der verwendeten Instrumente (Injektionsnadeln, elektrochirurgische Messer mit unterschiedlichen Konfigurationen, hämostatische Zangen) könnte wertvolle Informationen über den Fortschritt und die Verfahrensmerkmale der ESD liefern und eine automatische standardisierte Berichterstattung ermöglichen.
Ziele: Ziel dieser Studie war die Entwicklung eines KI-Algorithmus zur Erkennung und Delineation von endoskopischen Instrumenten bei der ESD.
Methodik: 17 ESD-Videos (9×rektal, 5×ösophageal, 3×gastrisch) wurden retrospektiv zusammengestellt. Auf 8530 Einzelbilder dieser Videos wurden durch 2 Studienmitarbeiter die folgenden Klassen eingezeichnet: Hakenmesser – Spitze, Hakenmesser – Katheter, Nadelmesser – Spitze und – Katheter, Injektionsnadel -Spitze und – Katheter sowie hämostatische Zange – Spitze und – Katheter. Der annotierte Datensatz wurde zum Training eines DeepLabV3+-Deep-Learning-Algorithmus mit ConvNeXt-Backbone zur Erkennung und Abgrenzung der genannten Klassen verwendet. Die Evaluation erfolgte durch 5-fache interne Kreuzvalidierung.
Ergebnis: Die Validierung auf Einzelpixelbasis ergab insgesamt einen F1-Score von 0,80, eine Sensitivität von 0,81 und eine Spezifität von 1,00. Es wurden F1-Scores von 1,00, 0,97, 0,80, 0,98, 0,85, 0,97, 0,80, 0,51 bzw. 0,85 für die Klassen Hakenmesser – Katheter und – Spitze, Nadelmesser – Katheter und – Spitze, Injektionsnadel – Katheter und – Spitze, hämostatische Zange – Katheter und – Spitze gemessen.
Schlussfolgerung: In dieser Studie wurden die wichtigsten endoskopischen Instrumente, die während der ESD verwendet werden, mit hoher Genauigkeit erkannt. Die geringere Leistung bei der hämostatische Zange – Katheter kann auf die Unterrepräsentation dieser Klassen in den Trainingsdaten zurückgeführt werden. Zukünftige Studien werden sich auf die Erweiterung der Instrumentenklassen sowie auf die Ausbalancierung der Trainingsdaten konzentrieren.
Reliable recognition and localization of surgical instruments in endoscopic video recordings are foundational for a wide range of applications in computer- and robot-assisted minimally invasive surgery (RAMIS), including surgical training, skill assessment, and autonomous assistance. However, robust performance under real-world conditions remains a significant challenge. Incorporating surgical context - such as the current procedural phase - has emerged as a promising strategy to improve robustness and interpretability. To address these challenges, we organized the Surgical Procedure Phase, Keypoint, and Instrument Recognition (PhaKIR) sub-challenge as part of the Endoscopic Vision (EndoVis) challenge at MICCAI 2024. We introduced a novel, multi-center dataset comprising thirteen full-length laparoscopic cholecystectomy videos collected from three distinct medical institutions, with unified annotations for three interrelated tasks: surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation. Unlike existing datasets, ours enables joint investigation of instrument localization and procedural context within the same data while supporting the integration of temporal information across entire procedures. We report results and findings in accordance with the BIAS guidelines for biomedical image analysis challenges. The PhaKIR sub-challenge advances the field by providing a unique benchmark for developing temporally aware, context-driven methods in RAMIS and offers a high-quality resource to support future research in surgical scene understanding.
The growing reliance on Artificial Intelligence (AI) in critical domains such as healthcare demands robust mechanisms to ensure the trustworthiness of these systems, especially when faced with unexpected or anomalous inputs. This paper introduces the Open Medical Imaging Benchmarks for Out-Of-Distribution Detection (OpenMIBOOD), a comprehensive framework for evaluating out-of-distribution (OOD) detection methods specifically in medical imaging contexts. OpenMIBOOD includes three benchmarks from diverse medical domains, encompassing 14 datasets divided into covariate-shifted in-distribution, nearOOD, and far-OOD categories. We evaluate 24 post-hoc methods across these benchmarks, providing a standardized reference to advance the development and fair comparison of OODdetection methods. Results reveal that findings from broad-scale OOD benchmarks in natural image domains do not translate to medical applications, underscoring the critical need for such benchmarks in the medical field. By mitigating the risk of exposing AI models to inputs outside their training distribution, OpenMIBOOD aims to support the advancement of reliable and trustworthy AI systems in healthcare. The repository is available at https://github.com/remic-othr/OpenMIBOOD.
Objective
Despite high stand-alone performance, studies demonstrate that artificial intelligence (AI)-supported endoscopic diagnostics often fall short in clinical applications due to human-AI interaction factors. This video-based trial on Barrett's esophagus aimed to investigate how examiner behavior, their levels of confidence, and system usability influence the diagnostic outcomes of AI-assisted endoscopy.
Methods
The present analysis employed data from a multicenter randomized controlled tandem video trial involving 22 endoscopists with varying degrees of expertise. Participants were tasked with evaluating a set of 96 endoscopic videos of Barrett's esophagus in two distinct rounds, with and without AI assistance. Diagnostic confidence levels were recorded, and decision changes were categorized according to the AI prediction. Additional surveys assessed user experience and system usability ratings.
Results
AI assistance significantly increased examiner confidence levels (p < 0.001) and accuracy. Withdrawing AI assistance decreased confidence (p < 0.001), but not accuracy. Experts consistently reported higher confidence than non-experts (p < 0.001), regardless of performance. Despite improved confidence, correct AI guidance was disregarded in 16% of all cases, and 9% of initially correct diagnoses were changed to incorrect ones. Overreliance on AI, algorithm aversion, and uncertainty in AI predictions were identified as key factors influencing outcomes. The System Usability Scale questionnaire scores indicated good to excellent usability, with non-experts scoring 73.5 and experts 85.6.
Conclusions
Our findings highlight the pivotal function of examiner behavior in AI-assisted endoscopy. To fully realize the benefits of AI, implementing explainable AI, improving user interfaces, and providing targeted training are essential. Addressing these factors could enhance diagnostic accuracy and confidence in clinical practice.