@article{KrenzerHeilFittingetal.,
  author    = {Krenzer, Adrian and Heil, Stefan and Fitting, Daniel and Matti, Safa and Zoller, Wolfram G. and Hann, Alexander and Puppe, Frank},
  title     = {Automated classification of polyps using deep learning architectures and few-shot learning},
  series = {BMC Medical Imaging},
  volume    = {23},
  journal   = {BMC Medical Imaging},
  doi       = {10.1186/s12880-023-01007-4},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-357465},
  abstract  = {Background Colorectal cancer is a leading cause of cancer-related deaths worldwide. The best method to prevent CRC is a colonoscopy. However, not all colon polyps have the risk of becoming cancerous. Therefore, polyps are classified using different classification systems. After the classification, further treatment and procedures are based on the classification of the polyp. Nevertheless, classification is not easy. Therefore, we suggest two novel automated classifications system assisting gastroenterologists in classifying polyps based on the NICE and Paris classification. Methods We build two classification systems. One is classifying polyps based on their shape (Paris). The other classifies polyps based on their texture and surface patterns (NICE). A two-step process for the Paris classification is introduced: First, detecting and cropping the polyp on the image, and secondly, classifying the polyp based on the cropped area with a transformer network. For the NICE classification, we design a few-shot learning algorithm based on the Deep Metric Learning approach. The algorithm creates an embedding space for polyps, which allows classification from a few examples to account for the data scarcity of NICE annotated images in our database. Results For the Paris classification, we achieve an accuracy of 89.35 \%, surpassing all papers in the literature and establishing a new state-of-the-art and baseline accuracy for other publications on a public data set. For the NICE classification, we achieve a competitive accuracy of 81.13 \% and demonstrate thereby the viability of the few-shot learning paradigm in polyp classification in data-scarce environments. Additionally, we show different ablations of the algorithms. Finally, we further elaborate on the explainability of the system by showing heat maps of the neural network explaining neural activations. Conclusion Overall we introduce two polyp classification systems to assist gastroenterologists. We achieve state-of-the-art performance in the Paris classification and demonstrate the viability of the few-shot learning paradigm in the NICE classification, addressing the prevalent data scarcity issues faced in medical machine learning.},
  language  = {en}
}
@article{LimanMayFetteetal.2023,
  author    = {Liman, Leon and May, Bernd and Fette, Georg and Krebs, Jonathan and Puppe, Frank},
  title     = {Using a clinical data warehouse to calculate and present key metrics for the radiology department: implementation and performance evaluation},
  series = {JMIR Medical Informatics},
  volume    = {11},
  journal   = {JMIR Medical Informatics},
  issn      = {2291-9694},
  doi       = {10.2196/41808},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-349411},
  year      = {2023},
  abstract  = {Background: Due to the importance of radiologic examinations, such as X-rays or computed tomography scans, for many clinical diagnoses, the optimal use of the radiology department is 1 of the primary goals of many hospitals. Objective: This study aims to calculate the key metrics of this use by creating a radiology data warehouse solution, where data from radiology information systems (RISs) can be imported and then queried using a query language as well as a graphical user interface (GUI). Methods: Using a simple configuration file, the developed system allowed for the processing of radiology data exported from any kind of RIS into a Microsoft Excel, comma-separated value (CSV), or JavaScript Object Notation (JSON) file. These data were then imported into a clinical data warehouse. Additional values based on the radiology data were calculated during this import process by implementing 1 of several provided interfaces. Afterward, the query language and GUI of the data warehouse were used to configure and calculate reports on these data. For the most common types of requested reports, a web interface was created to view their numbers as graphics. Results: The tool was successfully tested with the data of 4 different German hospitals from 2018 to 2021, with a total of 1,436,111 examinations. The user feedback was good, since all their queries could be answered if the available data were sufficient. The initial processing of the radiology data for using them with the clinical data warehouse took (depending on the amount of data provided by each hospital) between 7 minutes and 1 hour 11 minutes. Calculating 3 reports of different complexities on the data of each hospital was possible in 1-3 seconds for reports with up to 200 individual calculations and in up to 1.5 minutes for reports with up to 8200 individual calculations. Conclusions: A system was developed with the main advantage of being generic concerning the export of different RISs as well as concerning the configuration of queries for various reports. The queries could be configured easily using the GUI of the data warehouse, and their results could be exported into the standard formats Excel and CSV for further processing.},
  language  = {en}
}
@article{FischerHarteltPuppe2023,
  author    = {Fischer, Norbert and Hartelt, Alexander and Puppe, Frank},
  title     = {Line-level layout recognition of historical documents with background knowledge},
  series = {Algorithms},
  volume    = {16},
  journal   = {Algorithms},
  number    = {3},
  issn      = {1999-4893},
  doi       = {10.3390/a16030136},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-310938},
  year      = {2023},
  abstract  = {Digitization and transcription of historic documents offer new research opportunities for humanists and are the topics of many edition projects. However, manual work is still required for the main phases of layout recognition and the subsequent optical character recognition (OCR) of early printed documents. This paper describes and evaluates how deep learning approaches recognize text lines and can be extended to layout recognition using background knowledge. The evaluation was performed on five corpora of early prints from the 15th and 16th Centuries, representing a variety of layout features. While the main text with standard layouts could be recognized in the correct reading order with a precision and recall of up to 99.9\%, also complex layouts were recognized at a rate as high as 90\% by using background knowledge, the full potential of which was revealed if many pages of the same source were transcribed.},
  language  = {en}
}
@article{LuxBanckSassmannshausenetal.2022,
  author    = {Lux, Thomas J. and Banck, Michael and Saßmannshausen, Zita and Troya, Joel and Krenzer, Adrian and Fitting, Daniel and Sudarevic, Boban and Zoller, Wolfram G. and Puppe, Frank and Meining, Alexander and Hann, Alexander},
  title     = {Pilot study of a new freely available computer-aided polyp detection system in clinical practice},
  series = {International Journal of Colorectal Disease},
  volume    = {37},
  journal   = {International Journal of Colorectal Disease},
  number    = {6},
  doi       = {10.1007/s00384-022-04178-8},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-324459},
  pages     = {1349-1354},
  year      = {2022},
  abstract  = {Purpose Computer-aided polyp detection (CADe) systems for colonoscopy are already presented to increase adenoma detection rate (ADR) in randomized clinical trials. Those commercially available closed systems often do not allow for data collection and algorithm optimization, for example regarding the usage of different endoscopy processors. Here, we present the first clinical experiences of a, for research purposes publicly available, CADe system. Methods We developed an end-to-end data acquisition and polyp detection system named EndoMind. Examiners of four centers utilizing four different endoscopy processors used EndoMind during their clinical routine. Detected polyps, ADR, time to first detection of a polyp (TFD), and system usability were evaluated (NCT05006092). Results During 41 colonoscopies, EndoMind detected 29 of 29 adenomas in 66 of 66 polyps resulting in an ADR of 41.5\%. Median TFD was 130 ms (95\%-CI, 80-200 ms) while maintaining a median false positive rate of 2.2\% (95\%-CI, 1.7-2.8\%). The four participating centers rated the system using the System Usability Scale with a median of 96.3 (95\%-CI, 70-100). Conclusion EndoMind's ability to acquire data, detect polyps in real-time, and high usability score indicate substantial practical value for research and clinical practice. Still, clinical benefit, measured by ADR, has to be determined in a prospective randomized controlled trial.},
  language  = {en}
}
@article{DakroubVermaFuehringAgorastouetal.2022,
  author    = {Dakroub, Mohamad and Verma-Fuehring, Raoul and Agorastou, Vaia and Sch{\"o}n, Julian and Hillenkamp, Jost and Puppe, Frank and Loewen, Nils A.},
  title     = {Inter-eye correlation analysis of 24-h IOPs and glaucoma progression},
  series = {Graefe's Archive for Clinical and Experimental Ophthalmology},
  volume    = {260},
  journal   = {Graefe's Archive for Clinical and Experimental Ophthalmology},
  number    = {10},
  doi       = {10.1007/s00417-022-05651-4},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-323831},
  pages     = {3349-3356},
  year      = {2022},
  abstract  = {Purpose To determine whether 24-h IOP monitoring can be a predictor for glaucoma progression and to analyze the inter-eye relationship of IOP, perfusion, and progression parameters. Methods We extracted data from manually drawn IOP curves with HIOP-Reader, a software suite we developed. The relationship between measured IOPs and mean ocular perfusion pressures (MOPP) to retinal nerve fiber layer (RNFL) thickness was analyzed. We determined the ROC curves for peak IOP (T\(_{max}\)), average IOP(T\(_{avg}\)), IOP variation (IOP\(_{var}\)), and historical IOP cut-off levels to detect glaucoma progression (rate of RNFL loss). Bivariate analysis was also conducted to check for various inter-eye relationships. Results Two hundred seventeen eyes were included. The average IOP was 14.8 ± 3.5 mmHg, with a 24-h variation of 5.2 ± 2.9 mmHg. A total of 52\% of eyes with RNFL progression data showed disease progression. There was no significant difference in T\(_{max}\), T\(_{avg}\), and IOP\(_{var}\) between progressors and non-progressors (all p > 0.05). Except for T\(_{avg}\) and the temporal RNFL, there was no correlation between disease progression in any quadrant and T\(_{max}\), T\(_{avg}\), and IOP\(_{var}\). Twenty-four-hour and outpatient IOP variables had poor sensitivities and specificities in detecting disease progression. The correlation of inter-eye parameters was moderate; correlation with disease progression was weak. Conclusion In line with our previous study, IOP data obtained during a single visit (outpatient or inpatient monitoring) make for a poor diagnostic tool, no matter the method deployed. Glaucoma progression and perfusion pressure in left and right eyes correlated weakly to moderately with each other. Key messages What is known: ● Our prior study showed that manually obtained 24-hour inpatient IOP measurements in right eyes are poor predictors for glaucoma progression. The inter-eye relationship of 24-hour IOP parameters and disease progression on optical coherence tomography (OCT) has not been examined. What we found: ● 24-hour IOP profiles of left eyes from the same study were a poor diagnostic tool to detect worsening glaucoma. ● Significant inter-eye correlations of various strengths were found for all tested parameters},
  language  = {en}
}
@article{KempfKrugPuppe2023,
  author    = {Kempf, Sebastian and Krug, Markus and Puppe, Frank},
  title     = {KIETA: Key-insight extraction from scientific tables},
  series = {Applied Intelligence},
  volume    = {53},
  journal   = {Applied Intelligence},
  number    = {8},
  issn      = {0924-669X},
  doi       = {10.1007/s10489-022-03957-8},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-324180},
  pages     = {9513-9530},
  year      = {2023},
  abstract  = {An important but very time consuming part of the research process is literature review. An already large and nevertheless growing ground set of publications as well as a steadily increasing publication rate continue to worsen the situation. Consequently, automating this task as far as possible is desirable. Experimental results of systems are key-insights of high importance during literature review and usually represented in form of tables. Our pipeline KIETA exploits these tables to contribute to the endeavor of automation by extracting them and their contained knowledge from scientific publications. The pipeline is split into multiple steps to guarantee modularity as well as analyzability, and agnosticim regarding the specific scientific domain up until the knowledge extraction step, which is based upon an ontology. Additionally, a dataset of corresponding articles has been manually annotated with information regarding table and knowledge extraction. Experiments show promising results that signal the possibility of an automated system, while also indicating limits of extracting knowledge from tables without any context.},
  language  = {en}
}
@article{Puppe2022,
  author    = {Puppe, Frank},
  title     = {Gesellschaftliche Perspektiven einer fachspezifischen KI f{\"u}r automatisierte Entscheidungen},
  series = {Informatik Spektrum},
  volume    = {45},
  journal   = {Informatik Spektrum},
  number    = {2},
  issn      = {0170-6012},
  doi       = {10.1007/s00287-022-01443-6},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-324197},
  pages     = {88-95},
  year      = {2022},
  abstract  = {Die k{\"u}nstliche Intelligenz (KI) entwickelt sich rasant und hat bereits eindrucksvolle Erfolge zu verzeichnen, darunter {\"u}bermenschliche Kompetenz in den meisten Spielen und vielen Quizshows, intelligente Suchmaschinen, individualisierte Werbung, Spracherkennung, -ausgabe und -{\"u}bersetzung auf sehr hohem Niveau und hervorragende Leistungen bei der Bildverarbeitung, u. a. in der Medizin, der optischen Zeichenerkennung, beim autonomen Fahren, aber auch beim Erkennen von Menschen auf Bildern und Videos oder bei Deep Fakes f{\"u}r Fotos und Videos. Es ist zu erwarten, dass die KI auch in der Entscheidungsfindung Menschen {\"u}bertreffen wird; ein alter Traum der Expertensysteme, der durch Lernverfahren, Big Data und Zugang zu dem gesammelten Wissen im Web in greifbare N{\"a}he r{\"u}ckt. Gegenstand dieses Beitrags sind aber weniger die technischen Entwicklungen, sondern m{\"o}gliche gesellschaftliche Auswirkungen einer spezialisierten, kompetenten KI f{\"u}r verschiedene Bereiche der autonomen, d. h. nicht nur unterst{\"u}tzenden Entscheidungsfindung: als Fußballschiedsrichter, in der Medizin, f{\"u}r richterliche Entscheidungen und sehr spekulativ auch im politischen Bereich. Dabei werden Vor- und Nachteile dieser Szenarien aus gesellschaftlicher Sicht diskutiert.},
  subject      = {K{\"u}nstliche Intelligenz},
  language  = {de}
}
@article{KrenzerBanckMakowskietal.2023,
  author    = {Krenzer, Adrian and Banck, Michael and Makowski, Kevin and Hekalo, Amar and Fitting, Daniel and Troya, Joel and Sudarevic, Boban and Zoller, Wolfgang G. and Hann, Alexander and Puppe, Frank},
  title     = {A real-time polyp-detection system with clinical application in colonoscopy using deep convolutional neural networks},
  series = {Journal of Imaging},
  volume    = {9},
  journal   = {Journal of Imaging},
  number    = {2},
  issn      = {2313-433X},
  doi       = {10.3390/jimaging9020026},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-304454},
  year      = {2023},
  abstract  = {Colorectal cancer (CRC) is a leading cause of cancer-related deaths worldwide. The best method to prevent CRC is with a colonoscopy. During this procedure, the gastroenterologist searches for polyps. However, there is a potential risk of polyps being missed by the gastroenterologist. Automated detection of polyps helps to assist the gastroenterologist during a colonoscopy. There are already publications examining the problem of polyp detection in the literature. Nevertheless, most of these systems are only used in the research context and are not implemented for clinical application. Therefore, we introduce the first fully open-source automated polyp-detection system scoring best on current benchmark data and implementing it ready for clinical application. To create the polyp-detection system (ENDOMIND-Advanced), we combined our own collected data from different hospitals and practices in Germany with open-source datasets to create a dataset with over 500,000 annotated images. ENDOMIND-Advanced leverages a post-processing technique based on video detection to work in real-time with a stream of images. It is integrated into a prototype ready for application in clinical interventions. We achieve better performance compared to the best system in the literature and score a F1-score of 90.24\% on the open-source CVC-VideoClinicDB benchmark.},
  language  = {en}
}
@article{HarteltPuppe2022,
  author    = {Hartelt, Alexander and Puppe, Frank},
  title     = {Optical Medieval Music Recognition using background knowledge},
  series = {Algorithms},
  volume    = {15},
  journal   = {Algorithms},
  number    = {7},
  issn      = {1999-4893},
  doi       = {10.3390/a15070221},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-278756},
  year      = {2022},
  abstract  = {This paper deals with the effect of exploiting background knowledge for improving an OMR (Optical Music Recognition) deep learning pipeline for transcribing medieval, monophonic, handwritten music from the 12th-14th century, whose usage has been neglected in the literature. Various types of background knowledge about overlapping notes and text, clefs, graphical connections (neumes) and their implications on the position in staff of the notes were used and evaluated. Moreover, the effect of different encoder/decoder architectures and of different datasets for training a mixed model and for document-specific fine-tuning based on an extended OMR pipeline with an additional post-processing step were evaluated. The use of background models improves all metrics and in particular the melody accuracy rate (mAR), which is based on the insert, delete and replace operations necessary to convert the generated melody into the correct melody. When using a mixed model and evaluating on a different dataset, our best model achieves without fine-tuning and without post-processing a mAR of 90.4\%, which is raised by nearly 30\% to 93.2\% mAR using background knowledge. With additional fine-tuning, the contribution of post-processing is even greater: the basic mAR of 90.5\% is raised by more than 50\% to 95.8\% mAR.},
  language  = {en}
}
@article{KrenzerMakowskiHekaloetal.2022,
  author    = {Krenzer, Adrian and Makowski, Kevin and Hekalo, Amar and Fitting, Daniel and Troya, Joel and Zoller, Wolfram G. and Hann, Alexander and Puppe, Frank},
  title     = {Fast machine learning annotation in the medical domain: a semi-automated video annotation tool for gastroenterologists},
  series = {BioMedical Engineering OnLine},
  volume    = {21},
  journal   = {BioMedical Engineering OnLine},
  number    = {1},
  doi       = {10.1186/s12938-022-01001-x},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-300231},
  year      = {2022},
  abstract  = {Background Machine learning, especially deep learning, is becoming more and more relevant in research and development in the medical domain. For all the supervised deep learning applications, data is the most critical factor in securing successful implementation and sustaining the progress of the machine learning model. Especially gastroenterological data, which often involves endoscopic videos, are cumbersome to annotate. Domain experts are needed to interpret and annotate the videos. To support those domain experts, we generated a framework. With this framework, instead of annotating every frame in the video sequence, experts are just performing key annotations at the beginning and the end of sequences with pathologies, e.g., visible polyps. Subsequently, non-expert annotators supported by machine learning add the missing annotations for the frames in-between. Methods In our framework, an expert reviews the video and annotates a few video frames to verify the object's annotations for the non-expert. In a second step, a non-expert has visual confirmation of the given object and can annotate all following and preceding frames with AI assistance. After the expert has finished, relevant frames will be selected and passed on to an AI model. This information allows the AI model to detect and mark the desired object on all following and preceding frames with an annotation. Therefore, the non-expert can adjust and modify the AI predictions and export the results, which can then be used to train the AI model. Results Using this framework, we were able to reduce workload of domain experts on average by a factor of 20 on our data. This is primarily due to the structure of the framework, which is designed to minimize the workload of the domain expert. Pairing this framework with a state-of-the-art semi-automated AI model enhances the annotation speed further. Through a prospective study with 10 participants, we show that semi-automated annotation using our tool doubles the annotation speed of non-expert annotators compared to a well-known state-of-the-art annotation tool. Conclusion In summary, we introduce a framework for fast expert annotation for gastroenterologists, which reduces the workload of the domain expert considerably while maintaining a very high annotation quality. The framework incorporates a semi-automated annotation system utilizing trained object detection models. The software and framework are open-source.},
  language  = {en}
}