Philipp C. Müller
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Philipp C. Müller has a total of 486 co-authors. Here are the top co-authors by number of shared publications:
- Daniel Rueckert
Technical University of Munich
17 shared publications Felix Nickel
Klinik für Allgemein, Viszeral und Thoraxchirurgie Universitätsklinikum Hamburg-Eppendorf
10 shared publicationsChristoph Kuemmerli
Department of Surgery, Clarunis – University Centre for Gastrointestinal and Hepatopancreatobiliary Diseases, Basel, Switzerland
9 shared publicationsBeat P. Müller-Stich
Clarunis University Centre for Gastrointestinal and Liver Diseases, 4002 Basel, Switzerland
8 shared publications- Christoph Kuemmerli
Department of Surgery, Clarunis University Digestive Health Care Center, University Hospital Basel, Basel, Switzerland
8 shared publications - Georgios Kaissis
Technical University of Munich
7 shared publications
Latest Publications
- 25 September 2026
- Published by Springer Nature
- p. 148-157
Cross-Modal Contrastive Learning of ECG and Angiography Representations for Severe Stenosis Classification
- 25 September 2026
- Published by Springer Nature
- p. 148-157
Cross-Modal Contrastive Learning of ECG and Angiography Representations for Severe Stenosis Classification
- 25 September 2026
- Published by Springer Nature
- p. 185-195
Echo2ECG: Enhancing ECG Representations with Cardiac Morphology from Multi-view Echos
- 25 September 2026
- Published by Springer Nature
- p. 185-195
Echo2ECG: Enhancing ECG Representations with Cardiac Morphology from Multi-view Echos
- Open Access
- 18 June 2026
- Published by Wolters Kluwer Health in Annals of Surgery
Learning Curves and Complexity Evolution in Robotic Liver Surgery
Objective: To evaluate the institutional learning curve of robotic liver resection (RLR), its association with surgical complexity and outcomes in an international multicenter cohort. Summary of Background Data: RLR is increasingly adopted worldwide, but the dynamics of the learning process and its impact on surgical outcomes remain poorly defined. Methods: This international, multicenter, retrospective cohort study included consecutive patients who underwent RLR between 06/2013 and 12/2024 across 22 expert centers. Based on the first 150 cases at each center, three phases were used to define the learning curve: operative time for competency, major complications for proficiency, and textbook outcome for mastery. Risk-adjusted generalized additive models were used to assess the learning process. Benchmark performance was compared to international reference values. Results: Among 2,331 patients, the most common indications were hepatocellular carcinoma (36%), benign lesions (23%) and colorectal liver metastases (21%). Operative time decreased after achieving competency (case #62), from 240 to 220 minutes ( P P =0.144), while progression to mastery (cases #139–150) was associated with a higher rate of textbook outcome (76% vs. 87%, P =0.021). In contrast, conversion (5.1% vs. 4.9% vs. 2%), bile leak (4.0% vs. 5.3% vs. 3.8%), and 90-day mortality (0.7% vs. 0.9% vs. 0.9%) remained low and stable across all learning phases. Importantly, benchmark criteria for operative time, hospital stay, and textbook outcome were met across all complexity groups, including during the early learning phase. Conclusions: Robotic liver surgery shows progressive outcome improvement with early achievement of benchmark-level results and stable performance across increasing complexity. These findings indicate that, in experienced centers, RLR can be safely implemented and integrated into clinical practice, supporting its key role in modern hepatobiliary surgery.
- 18 June 2026
- Published by Wolters Kluwer Health in Annals of Surgery
Open AccessLearning Curves and Complexity Evolution in Robotic Liver Surgery
Objective: To evaluate the institutional learning curve of robotic liver resection (RLR), its association with surgical complexity and outcomes in an international multicenter cohort. Summary of Background Data: RLR is increasingly adopted worldwide, but the dynamics of the learning process and its impact on surgical outcomes remain poorly defined. Methods: This international, multicenter, retrospective cohort study included consecutive patients who underwent RLR between 06/2013 and 12/2024 across 22 expert centers. Based on the first 150 cases at each center, three phases were used to define the learning curve: operative time for competency, major complications for proficiency, and textbook outcome for mastery. Risk-adjusted generalized additive models were used to assess the learning process. Benchmark performance was compared to international reference values. Results: Among 2,331 patients, the most common indications were hepatocellular carcinoma (36%), benign lesions (23%) and colorectal liver metastases (21%). Operative time decreased after achieving competency (case #62), from 240 to 220 minutes ( P P =0.144), while progression to mastery (cases #139–150) was associated with a higher rate of textbook outcome (76% vs. 87%, P =0.021). In contrast, conversion (5.1% vs. 4.9% vs. 2%), bile leak (4.0% vs. 5.3% vs. 3.8%), and 90-day mortality (0.7% vs. 0.9% vs. 0.9%) remained low and stable across all learning phases. Importantly, benchmark criteria for operative time, hospital stay, and textbook outcome were met across all complexity groups, including during the early learning phase. Conclusions: Robotic liver surgery shows progressive outcome improvement with early achievement of benchmark-level results and stable performance across increasing complexity. These findings indicate that, in experienced centers, RLR can be safely implemented and integrated into clinical practice, supporting its key role in modern hepatobiliary surgery.
- Open Access
- 25 March 2026
- Published by Wolters Kluwer Health in Annals of Surgery Open
- Vol. 7 (2) , e656
Minimally Invasive Versus Open Pancreatoduodenectomy: A Systematic Review and Meta-Analysis of Randomized Controlled Trials
Objective: To compare perioperative outcomes of minimally invasive pancreatoduodenectomy (MIPD) to open pancreatoduodenectomy (OPD) using evidence from randomized controlled trials (RCTs). Background: The wider adoption of MIPD has largely been fueled by observational studies rather than high-level evidence. Methods: We searched Cochrane Central Register of Controlled Trials, MEDLINE, and Web of Science for RCTs comparing MIPD with OPD in adult patients with benign or malignant conditions requiring elective pancreatoduodenectomy. The primary outcomes were 90-day mortality, the comprehensive complication index, Clavien-Dindo grade ≥III complications, and hospital length of stay (LOS). Secondary outcomes included postoperative pancreatic fistula (POPF), delayed gastric emptying (DGE), postpancreatectomy hemorrhage (PPH), blood loss, reoperation, operative time, and oncologic outcomes. Data were pooled as odds ratios or mean differences using a random-effects model. Risk of bias was assessed using the Cochrane risk of bias tool, and the certainty of evidence was evaluated according to the Grading of Recommendations Assessment, Development and Evaluation approach (PROSPERO ID: CRD42024592919). Results: Ten RCTs with a total of 1794 patients were included. Meta-analysis showed there were no significant differences regarding 90-day mortality, Clavien-Dindo ≥3 complications, POPF, DGE, PPH, reoperation, readmission, or oncologic outcomes between MIPD and OPD. LOS was reduced for MIPD. No clinically relevant differences were found in the subgroup analyses of laparoscopic and robotic pancreatoduodenectomy. Certainty of evidence was moderate to low. Conclusions: MIPD showed no clinically relevant advantages over OPD. These findings were consistent both for the robotic and laparoscopic approach.
- 25 March 2026
- Published by Wolters Kluwer Health in Annals of Surgery Open
- Vol. 7 (2) , e656
Open AccessMinimally Invasive Versus Open Pancreatoduodenectomy: A Systematic Review and Meta-Analysis of Randomized Controlled Trials
Objective: To compare perioperative outcomes of minimally invasive pancreatoduodenectomy (MIPD) to open pancreatoduodenectomy (OPD) using evidence from randomized controlled trials (RCTs). Background: The wider adoption of MIPD has largely been fueled by observational studies rather than high-level evidence. Methods: We searched Cochrane Central Register of Controlled Trials, MEDLINE, and Web of Science for RCTs comparing MIPD with OPD in adult patients with benign or malignant conditions requiring elective pancreatoduodenectomy. The primary outcomes were 90-day mortality, the comprehensive complication index, Clavien-Dindo grade ≥III complications, and hospital length of stay (LOS). Secondary outcomes included postoperative pancreatic fistula (POPF), delayed gastric emptying (DGE), postpancreatectomy hemorrhage (PPH), blood loss, reoperation, operative time, and oncologic outcomes. Data were pooled as odds ratios or mean differences using a random-effects model. Risk of bias was assessed using the Cochrane risk of bias tool, and the certainty of evidence was evaluated according to the Grading of Recommendations Assessment, Development and Evaluation approach (PROSPERO ID: CRD42024592919). Results: Ten RCTs with a total of 1794 patients were included. Meta-analysis showed there were no significant differences regarding 90-day mortality, Clavien-Dindo ≥3 complications, POPF, DGE, PPH, reoperation, readmission, or oncologic outcomes between MIPD and OPD. LOS was reduced for MIPD. No clinically relevant differences were found in the subgroup analyses of laparoscopic and robotic pancreatoduodenectomy. Certainty of evidence was moderate to low. Conclusions: MIPD showed no clinically relevant advantages over OPD. These findings were consistent both for the robotic and laparoscopic approach.
- Open Access
- 9 February 2026
- Published by Springer Nature in Surgical Endoscopy
- Vol. 40 (4) , 3389-3402
Clinical practice of indocyanine green fluorescence imaging in robotic liver surgery – a global expert survey
Background Indocyanine green (ICG) fluorescence imaging is increasingly incorporated into robotic liver resections (RLR), yet clinical practice regarding timing, dosage, and staining techniques is divergent. This international expert survey aimed to characterize current practices for ICG in RLR. Methods Experts in RLR were invited to participate based on surgical volume (experience of \(\ge\) 50 RLR and \(\ge\) 30 annual RLR). A 74-item questionnaire was developed following a literature search and reviewed by a steering committee. The survey addressed indications, timing, dosage, imaging technology, benefits, limitations, training, and future directions of ICG use. Responses collected between September and October 2025 were analyzed. Results Seventy experts from 19 countries completed the survey, corresponding to an 88% response rate. Centers performed a median of 180 annual liver resections, including 55 RLR. Most experts used ICG (96%) during RLR. Anatomical demarcation (91%), tumor localization (60%), and biliary anatomy assessment (60%) were the most frequent indications. 60% of experts use preoperative ICG, while intraoperative ICG is mainly administered for demarcation (67%) and biliary tract visualization (40%). Considerable heterogeneity exists in dosage, timing, and staining techniques, particularly in cirrhotic livers and for tumor localization. 53% of the experts had standard operating procedures, whereas 64% expressed the need for a higher degree of standardization. Reported benefits of ICG use included improved anatomical orientation, margin assessment, lesion detection, and support during complex resections. Perceived limitations included background fluorescence, tissue penetration and variable staining in diseased parenchyma. 80% anticipated improved outcomes with combined ICG and three-dimensional image-guidance. Conclusion ICG fluorescence is widely used in RLR and is an important cornerstone for precision-guided robotic liver surgery. Standardized clinical practice guidelines, structured training, and technological improvements in imaging and navigation systems are claimed to optimize its clinical use.- 9 February 2026
- Published by Springer Nature in Surgical Endoscopy
- Vol. 40 (4) , 3389-3402
Open AccessClinical practice of indocyanine green fluorescence imaging in robotic liver surgery – a global expert survey
Background Indocyanine green (ICG) fluorescence imaging is increasingly incorporated into robotic liver resections (RLR), yet clinical practice regarding timing, dosage, and staining techniques is divergent. This international expert survey aimed to characterize current practices for ICG in RLR. Methods Experts in RLR were invited to participate based on surgical volume (experience of \(\ge\) 50 RLR and \(\ge\) 30 annual RLR). A 74-item questionnaire was developed following a literature search and reviewed by a steering committee. The survey addressed indications, timing, dosage, imaging technology, benefits, limitations, training, and future directions of ICG use. Responses collected between September and October 2025 were analyzed. Results Seventy experts from 19 countries completed the survey, corresponding to an 88% response rate. Centers performed a median of 180 annual liver resections, including 55 RLR. Most experts used ICG (96%) during RLR. Anatomical demarcation (91%), tumor localization (60%), and biliary anatomy assessment (60%) were the most frequent indications. 60% of experts use preoperative ICG, while intraoperative ICG is mainly administered for demarcation (67%) and biliary tract visualization (40%). Considerable heterogeneity exists in dosage, timing, and staining techniques, particularly in cirrhotic livers and for tumor localization. 53% of the experts had standard operating procedures, whereas 64% expressed the need for a higher degree of standardization. Reported benefits of ICG use included improved anatomical orientation, margin assessment, lesion detection, and support during complex resections. Perceived limitations included background fluorescence, tissue penetration and variable staining in diseased parenchyma. 80% anticipated improved outcomes with combined ICG and three-dimensional image-guidance. Conclusion ICG fluorescence is widely used in RLR and is an important cornerstone for precision-guided robotic liver surgery. Standardized clinical practice guidelines, structured training, and technological improvements in imaging and navigation systems are claimed to optimize its clinical use. - Open Access
Multi-View Stenosis Classification Leveraging Transformer-Based Multiple-Instance Learning Using Real-World Clinical Data
Coronary artery stenosis is a leading cause of cardiovascular disease, diagnosed by analyzing the coronary arteries from multiple angiography views. Although numerous deep-learning models have been proposed for stenosis detection from a single angiography view, their performance heavily relies on expensive view-level annotations, which are often not readily available in hospital systems. Moreover, these models fail to capture the temporal dynamics and dependencies among multiple views, which are crucial for clinical diagnosis. To address this, we propose SegmentMIL, a transformer-based multi-view multiple-instance learning framework for patient-level stenosis classification. Trained on a real-world clinical dataset, using patient-level supervision and without any view-level annotations, SegmentMIL jointly predicts the presence of stenosis and localizes the affected anatomical region, distinguishing between the right and left coronary arteries and their respective segments. SegmentMIL obtains high performance on internal and external evaluations and outperforms both view-level models and classical MIL baselines, underscoring its potential as a clinically viable and scalable solution for coronary stenosis diagnosis. Our code is available at https://github.com/NikolaCenic/mil-stenosis.
Open AccessMulti-View Stenosis Classification Leveraging Transformer-Based Multiple-Instance Learning Using Real-World Clinical Data
Coronary artery stenosis is a leading cause of cardiovascular disease, diagnosed by analyzing the coronary arteries from multiple angiography views. Although numerous deep-learning models have been proposed for stenosis detection from a single angiography view, their performance heavily relies on expensive view-level annotations, which are often not readily available in hospital systems. Moreover, these models fail to capture the temporal dynamics and dependencies among multiple views, which are crucial for clinical diagnosis. To address this, we propose SegmentMIL, a transformer-based multi-view multiple-instance learning framework for patient-level stenosis classification. Trained on a real-world clinical dataset, using patient-level supervision and without any view-level annotations, SegmentMIL jointly predicts the presence of stenosis and localizes the affected anatomical region, distinguishing between the right and left coronary arteries and their respective segments. SegmentMIL obtains high performance on internal and external evaluations and outperforms both view-level models and classical MIL baselines, underscoring its potential as a clinically viable and scalable solution for coronary stenosis diagnosis. Our code is available at https://github.com/NikolaCenic/mil-stenosis.
- Open Access
- 22 January 2026
- Published by Elsevier in JHEP Reports
- Vol. 8 (3) , 101732
HCC recurrence in liver transplants treated with hypothermic oxygenated machine perfusion: An international matched cohort study
- 22 January 2026
- Published by Elsevier in JHEP Reports
- Vol. 8 (3) , 101732
Open AccessHCC recurrence in liver transplants treated with hypothermic oxygenated machine perfusion: An international matched cohort study
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (8) , 1
Multi-View Stenosis Classification Leveraging Transformer-Based Multiple-Instance Learning Using Real-World Clinical Data
Coronary artery stenosis is a leading cause of cardiovascular disease, diagnosed by analyzing the coronary arteries from multiple angiography views. Although numerous deep-learning models have been proposed for stenosis detection from a single angiography view, their performance heavily relies on expensive view-level annotations, which are often not readily available in hospital systems. Moreover, these models fail to capture the temporal dynamics and dependencies among multiple views, which are crucial for clinical diagnosis. To address this, we propose SegmentMIL, a transformer-basedmulti-viewmultiple-instance learning framework for patient-level stenosis classification. Trained on a real-world clinical dataset, using patient-level supervision and without any view-level annotations, SegmentMIL jointly predicts the presence of stenosis and localizes the affected anatomical region, distinguishing between the right and left coronary arteries and their respective segments. SegmentMIL obtains high performance on internal and external evaluations and outperforms both view-level models and classical MIL baselines, underscoring its potential as a clinically viable and scalable solution for coronary stenosis diagnosis. Our code is available at https://github.com/NikolaCenic/mil-stenosis.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (8) , 1
Open AccessMulti-View Stenosis Classification Leveraging Transformer-Based Multiple-Instance Learning Using Real-World Clinical Data
Coronary artery stenosis is a leading cause of cardiovascular disease, diagnosed by analyzing the coronary arteries from multiple angiography views. Although numerous deep-learning models have been proposed for stenosis detection from a single angiography view, their performance heavily relies on expensive view-level annotations, which are often not readily available in hospital systems. Moreover, these models fail to capture the temporal dynamics and dependencies among multiple views, which are crucial for clinical diagnosis. To address this, we propose SegmentMIL, a transformer-basedmulti-viewmultiple-instance learning framework for patient-level stenosis classification. Trained on a real-world clinical dataset, using patient-level supervision and without any view-level annotations, SegmentMIL jointly predicts the presence of stenosis and localizes the affected anatomical region, distinguishing between the right and left coronary arteries and their respective segments. SegmentMIL obtains high performance on internal and external evaluations and outperforms both view-level models and classical MIL baselines, underscoring its potential as a clinically viable and scalable solution for coronary stenosis diagnosis. Our code is available at https://github.com/NikolaCenic/mil-stenosis.
- Open Access
- 29 December 2025
- Published by Oxford University Press (OUP) in BJS Open
- Vol. 10 (1)
Assessing complications following pancreatoduodenectomy: the Comprehensive Complication Index versus the Clavien–Dindo classification
Background: This study aimed to compare the accuracy of the Comprehensive Complication Index (CCI) with that of the Clavien–Dindo classification in patients undergoing pancreatoduodenectomy. Methods: A two-centre, retrospective study was undertaken that included patients who underwent pancreatoduodenectomy between 2008 and 2022. Three approaches were used to assess the two complication scores: the Spearman rank test, yielding the correlation coefficient (r), the area under the curve with 95% confidence intervals, and a mixed-effects model and a generalized mixed-effects model that yielded odds ratios and β-coefficients. Results: A total of 596 patients were included. The CCI and Clavien–Dindo classification demonstrated no correlation with 90-day mortality (r = − 0.021, 0.618; and r = −0.003, P = 0.951) but a significant correlation with length of hospital stay (r = 0.620, P < 0.001; and r = 0.605, P < 0.001) and with 90-day readmission rate (r = 0.148, P < 0.001; and r = 0.120, P = 0.005). The accuracy of the CCI was superior to that of the Clavien–Dindo classification for length of hospital stay dichotomized at the 75th (P = 0.022) and 90th (P < 0.001) percentiles. The CCI significantly improved the effect of the Clavien–Dindo classification (random effect, P < 0.001) in the mixed-effects and generalized mixed-effects logistic regression analyses. Conclusion: Compared with the Clavien–Dindo classification, the CCI appeared to be more accurate in terms of its association with a prolonged hospital stay and 90-day readmission rate. The CCI should complement the Clavien–Dindo classification in clinical and research settings.
- 29 December 2025
- Published by Oxford University Press (OUP) in BJS Open
- Vol. 10 (1)
Open AccessAssessing complications following pancreatoduodenectomy: the Comprehensive Complication Index versus the Clavien–Dindo classification
Background: This study aimed to compare the accuracy of the Comprehensive Complication Index (CCI) with that of the Clavien–Dindo classification in patients undergoing pancreatoduodenectomy. Methods: A two-centre, retrospective study was undertaken that included patients who underwent pancreatoduodenectomy between 2008 and 2022. Three approaches were used to assess the two complication scores: the Spearman rank test, yielding the correlation coefficient (r), the area under the curve with 95% confidence intervals, and a mixed-effects model and a generalized mixed-effects model that yielded odds ratios and β-coefficients. Results: A total of 596 patients were included. The CCI and Clavien–Dindo classification demonstrated no correlation with 90-day mortality (r = − 0.021, 0.618; and r = −0.003, P = 0.951) but a significant correlation with length of hospital stay (r = 0.620, P < 0.001; and r = 0.605, P < 0.001) and with 90-day readmission rate (r = 0.148, P < 0.001; and r = 0.120, P = 0.005). The accuracy of the CCI was superior to that of the Clavien–Dindo classification for length of hospital stay dichotomized at the 75th (P = 0.022) and 90th (P < 0.001) percentiles. The CCI significantly improved the effect of the Clavien–Dindo classification (random effect, P < 0.001) in the mixed-effects and generalized mixed-effects logistic regression analyses. Conclusion: Compared with the Clavien–Dindo classification, the CCI appeared to be more accurate in terms of its association with a prolonged hospital stay and 90-day readmission rate. The CCI should complement the Clavien–Dindo classification in clinical and research settings.