@misc{ZellmerRauberProbstetal., author = {Zellmer, Stephan and Rauber, David and Probst, Andreas and Weber, Tobias and Braun, Georg and Nagl, Sandra and R{\"o}mmele, Christoph and Schnoy, Elisabeth and Birzle, Lisa and Aehling, Niklas and Schulz, Dominik Andreas Helmut Otto and Palm, Christoph and Messmann, Helmut and Ebigbo, Alanna}, title = {K{\"u}nstliche Intelligenz als Hilfsmittel zur Detektion der Papilla duodeni major und des papill{\"a}ren Ostiums w{\"a}hrend der ERCP}, series = {Zeitschrift f{\"u}r Gastroenterologie}, volume = {63}, journal = {Zeitschrift f{\"u}r Gastroenterologie}, number = {5}, publisher = {Thieme}, address = {Stuttgart}, doi = {10.1055/s-0045-1806882}, pages = {e295}, abstract = {Einleitung Die Endoskopische Retrograde Cholangiopankreatikographie (ERCP) ist der Goldstandard in der endoskopischen Therapie von Erkrankungen des pankreatobili{\"a}ren Trakts. Allerdings ist sie technisch anspruchsvoll, schwer zu erlernen und mit einer relativ hohen Komplikationsrate assoziiert. Daher soll in der vorliegenden Machbarkeitsstudie gepr{\"u}ft werden, ob mithilfe eines Deeplearning- Algorithmus die Papille und das Ostium zuverl{\"a}ssig detektiert werden k{\"o}nnen und dieser f{\"u}r Endoskopiker, insbesondere in der Ausbildungssituation, ein geeignetes Hilfsmittel darstellen k{\"o}nnte. Material und Methodik Insgesamt wurden 1534 ERCP-Bilder von 134 Patienten analysiert, wobei sowohl die Papilla duodeni major als auch das Ostium segmentiert wurden. Anschließend erfolgte das Training eines neuronalen Netzes unter Verwendung eines Deep-Learning-Algorithmus. F{\"u}r den Test des Algorithmus erfolgte eine f{\"u}nffache Kreuzvalidierung. Ergebnisse Auf den 1534 gelabelten Bildern wurden f{\"u}r die Klasse Papille ein F1-Wert von 0,7996, eine Sensitivit{\"a}t von 0,8488 und eine Spezifit{\"a}t von 0,9822 erzielt. F{\"u}r die Klasse Ostium ergaben sich ein F1-Wert von 0,5198, eine Sensitivit{\"a}t von 0,5945 und eine Spezifit{\"a}t von 0,9974. Klassen{\"u}bergreifend (Klasse Papille und Klasse Ostium) betrug der F1-Wert 0,6593, die Sensitivit{\"a}t 0,7216 und f{\"u}r die Spezifit{\"a}t 0,9898. Zusammenfassung In der vorliegenden Machbarkeitsstudie zeigte das neuronale Netz eine hohe Sensitivit{\"a}t und eine sehr hohe Spezifit{\"a}t bei der Identifikation der Papilla duodeni major. Die Detektion des Ostiums erfolgte hingegen mit einer deutlich geringeren Sensitivit{\"a}t. Zuk{\"u}nftig ist eine Erweiterung des Trainingsdatensatzes um Videos und klinische Daten vorgesehen, um die Leistungsf{\"a}higkeit des Netzwerks zu verbessern. Hierdurch k{\"o}nnte langfristig ein geeignetes Assistenzsystem f{\"u}r die ERCP, insbesondere in der Ausbildungssituation etabliert werden.}, language = {de} } @article{MaerklRueckertRauberetal., author = {Maerkl, Raphaela and Rueckert, Tobias and Rauber, David and Gutbrod, Max and Weber Nunes, Danilo and Palm, Christoph}, title = {Enhancing generalization in zero-shot multi-label endoscopic instrument classification}, series = {International Journal of Computer Assisted Radiology and Surgery}, volume = {20}, journal = {International Journal of Computer Assisted Radiology and Surgery}, publisher = {Springer Nature}, doi = {10.1007/s11548-025-03439-5}, url = {http://nbn-resolving.de/urn:nbn:de:bvb:898-opus4-85674}, pages = {1577 -- 1587}, abstract = {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.}, language = {en} } @unpublished{RueckertRauberMaerkletal., author = {R{\"u}ckert, Tobias and Rauber, David and Maerkl, Raphaela and Klausmann, Leonard and Yildiran, Suemeyye R. and Gutbrod, Max and Nunes, Danilo Weber and Moreno, Alvaro Fernandez and Luengo, Imanol and Stoyanov, Danail and Toussaint, Nicolas and Cho, Enki and Kim, Hyeon Bae and Choo, Oh Sung and Kim, Ka Young and Kim, Seong Tae and Arantes, Gon{\c{c}}alo and Song, Kehan and Zhu, Jianjun and Xiong, Junchen and Lin, Tingyi and Kikuchi, Shunsuke and Matsuzaki, Hiroki and Kouno, Atsushi and Manesco, Jo{\~a}o Renato Ribeiro and Papa, Jo{\~a}o Paulo and Choi, Tae-Min and Jeong, Tae Kyeong and Park, Juyoun and Alabi, Oluwatosin and Wei, Meng and Vercauteren, Tom and Wu, Runzhi and Xu, Mengya and an Wang, and Bai, Long and Ren, Hongliang and Yamlahi, Amine and Hennighausen, Jakob and Maier-Hein, Lena and Kondo, Satoshi and Kasai, Satoshi and Hirasawa, Kousuke and Yang, Shu and Wang, Yihui and Chen, Hao and Rodr{\´i}guez, Santiago and Aparicio, Nicol{\´a}s and Manrique, Leonardo and Lyons, Juan Camilo and Hosie, Olivia and Ayobi, Nicol{\´a}s and Arbel{\´a}ez, Pablo and Li, Yiping and Khalil, Yasmina Al and Nasirihaghighi, Sahar and Speidel, Stefanie and R{\"u}ckert, Daniel and Feussner, Hubertus and Wilhelm, Dirk and Palm, Christoph}, title = {Comparative validation of surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation in endoscopy: Results of the PhaKIR 2024 challenge}, pages = {36}, abstract = {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.}, language = {en} } @misc{ZellmerRauberProbstetal., author = {Zellmer, Stephan and Rauber, David and Probst, Andreas and Weber, Tobias and Nagl, Sandra and R{\"o}mmele, Christoph and Schnoy, Elisabeth and Palm, Christoph and Messmann, Helmut and Ebigbo, Alanna}, title = {Verwendung k{\"u}nstlicher Intelligenz bei der Detektion der Papilla duodeni major}, series = {Zeitschrift f{\"u}r Gastroenterologie}, volume = {61}, journal = {Zeitschrift f{\"u}r Gastroenterologie}, number = {08}, publisher = {Thieme}, address = {Stuttgart}, doi = {10.1055/s-0043-1772000}, pages = {e539 -- e540}, abstract = {Einleitung Die Endoskopische Retrograde Cholangiopankreatikographie (ERCP) ist der Goldstandard in der Diagnostik und Therapie von Erkrankungen des pankreatobili{\"a}ren Trakts. Jedoch ist sie technisch sehr anspruchsvoll und weist eine vergleichsweise hohe Komplikationsrate auf. Ziele In der vorliegenden Machbarkeitsstudie soll gepr{\"u}ft werden, ob mithilfe eines Deep-learning-Algorithmus die Papille und das Ostium zuverl{\"a}ssig detektiert werden k{\"o}nnen und somit f{\"u}r Endoskopiker mit geringer Erfahrung ein geeignetes Hilfsmittel, insbesondere f{\"u}r die Ausbildungssituation, darstellen k{\"o}nnten. Methodik Wir betrachteten insgesamt 606 Bilddatens{\"a}tze von 65 Patienten. In diesen wurde sowohl die Papilla duodeni major als auch das Ostium segmentiert. Anschließend wurde eine neuronales Netz mittels eines Deep-learning-Algorithmus trainiert. Außerdem erfolgte eine 5-fache Kreuzvaldierung. Ergebnisse Bei einer 5-fachen Kreuzvaldierung auf den 606 gelabelten Daten konnte f{\"u}r die Klasse Papille eine F1-Wert von 0,7908, eine Sensitivit{\"a}t von 0,7943 und eine Spezifit{\"a}t von 0,9785 erreicht werden, f{\"u}r die Klasse Ostium eine F1-Wert von 0,5538, eine Sensitivit{\"a}t von 0,5094 und eine Spezifit{\"a}t von 0,9970 (vgl. [Tab. 1]). Unabh{\"a}ngig von der Klasse zeigte sich gemittelt (Klasse Papille und Klasse Ostium) ein F1-Wert von 0,6673, eine Sensitivit{\"a}t von 0,6519 und eine Spezifit{\"a}t von 0,9877 (vgl. [Tab. 2]). Schlussfolgerung In vorliegende Machbarkeitsstudie konnte das neuronale Netz die Papilla duodeni major mit einer hohen Sensitivit{\"a}t und sehr hohen Spezifit{\"a}t identifizieren. Bei der Detektion des Ostiums war die Sensitivit{\"a}t deutlich geringer. Zuk{\"u}nftig soll das das neuronale Netz mit mehr Daten trainiert werden. Außerdem ist geplant, den Algorithmus auch auf Videos anzuwenden. Somit k{\"o}nnte langfristig ein geeignetes Hilfsmittel f{\"u}r die ERCP etabliert werden.}, language = {de} } @article{RueckertRauberMaerkletal., author = {Rueckert, Tobias and Rauber, David and Maerkl, Raphaela and Klausmann, Leonard and Yildiran, Suemeyye R. and Gutbrod, Max and Nunes, Danilo Weber and Moreno, Alvaro Fernandez and Luengo, Imanol and Stoyanov, Danail and Toussaint, Nicolas and Cho, Enki and Kim, Hyeon Bae and Choo, Oh Sung and Kim, Ka Young and Kim, Seong Tae and Arantes, Gon{\c{c}}alo and Song, Kehan and Zhu, Jianjun and Xiong, Junchen and Lin, Tingyi and Kikuchi, Shunsuke and Matsuzaki, Hiroki and Kouno, Atsushi and Manesco, Jo{\~a}o Renato Ribeiro and Papa, Jo{\~a}o Paulo and Choi, Tae-Min and Jeong, Tae Kyeong and Park, Juyoun and Alabi, Oluwatosin and Wei, Meng and Vercauteren, Tom and Wu, Runzhi and Xu, Mengya and Wang, An and Bai, Long and Ren, Hongliang and Yamlahi, Amine and Hennighausen, Jakob and Maier-Hein, Lena and Kondo, Satoshi and Kasai, Satoshi and Hirasawa, Kousuke and Yang, Shu and Wang, Yihui and Chen, Hao and Rodr{\´i}guez, Santiago and Aparicio, Nicol{\´a}s and Manrique, Leonardo and Palm, Christoph and Wilhelm, Dirk and Feussner, Hubertus and Rueckert, Daniel and Speidel, Stefanie and Nasirihaghighi, Sahar and Al Khalil, Yasmina and Li, Yiping and Arbel{\´a}ez, Pablo and Ayobi, Nicol{\´a}s and Hosie, Olivia and Lyons, Juan Camilo}, title = {Comparative validation of surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation in endoscopy: Results of the PhaKIR 2024 challenge}, series = {Medical Image Analysis}, volume = {109}, journal = {Medical Image Analysis}, publisher = {Elsevier}, issn = {1361-8415}, doi = {10.1016/j.media.2026.103945}, pages = {31}, abstract = {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.}, language = {en} }