Christoph M. Friedrich

Matthias Eisenmann, Annika Reinke, Vivienn Weru et al. · utoronto

The number of international benchmarking competitions is steadily increasing in various fields of machine learning (ML) research and practice. So far, however, little is known about the common practice as well as bottlenecks faced by the community in tackling the research questions posed. To shed light on the status quo of algorithm development in the specific field of biomedical imaging analysis, we designed an international survey that was issued to all participants of challenges conducted in conjunction with the IEEE ISBI 2021 and MICCAI 2021 conferences (80 competitions in total). The survey covered participants' expertise and working environments, their chosen strategies, as well as algorithm characteristics. A median of 72% challenge participants took part in the survey. According to our results, knowledge exchange was the primary incentive (70%) for participation, while the reception of prize money played only a minor role (16%). While a median of 80 working hours was spent on method development, a large portion of participants stated that they did not have enough time for method development (32%). 25% perceived the infrastructure to be a bottleneck. Overall, 94% of all solutions were deep learning-based. Of these, 84% were based on standard architectures. 43% of the respondents reported that the data samples (e.g., images) were too large to be processed at once. This was most commonly addressed by patch-based training (69%), downsampling (37%), and solving 3D analysis tasks as a series of 2D tasks. K-fold cross-validation on the training set was performed by only 37% of the participants and only 50% of the participants performed ensembling based on multiple identical models (61%) or heterogeneous models (39%). 48% of the respondents applied postprocessing steps.

Amin Dada, Aokun Chen, Cheng Peng et al.

Traditionally, large language models have been either trained on general web crawls or domain-specific data. However, recent successes of generative large language models, have shed light on the benefits of cross-domain datasets. To examine the significance of prioritizing data diversity over quality, we present a German dataset comprising texts from five domains, along with another dataset aimed at containing high-quality data. Through training a series of models ranging between 122M and 750M parameters on both datasets, we conduct a comprehensive benchmark on multiple downstream tasks. Our findings demonstrate that the models trained on the cross-domain dataset outperform those trained on quality data alone, leading to improvements up to $4.45\%$ over the previous state-of-the-art. The models are available at https://huggingface.co/ikim-uk-essen

Matthias Eisenmann, Annika Reinke, Vivienn Weru et al.

International benchmarking competitions have become fundamental for the comparative performance assessment of image analysis methods. However, little attention has been given to investigating what can be learnt from these competitions. Do they really generate scientific progress? What are common and successful participation strategies? What makes a solution superior to a competing method? To address this gap in the literature, we performed a multi-center study with all 80 competitions that were conducted in the scope of IEEE ISBI 2021 and MICCAI 2021. Statistical analyses performed based on comprehensive descriptions of the submitted algorithms linked to their rank as well as the underlying participation strategies revealed common characteristics of winning solutions. These typically include the use of multi-task learning (63%) and/or multi-stage pipelines (61%), and a focus on augmentation (100%), image preprocessing (97%), data curation (79%), and postprocessing (66%). The "typical" lead of a winning team is a computer scientist with a doctoral degree, five years of experience in biomedical image analysis, and four years of experience in deep learning. Two core general development strategies stood out for highly-ranked teams: the reflection of the metrics in the method design and the focus on analyzing and handling failure cases. According to the organizers, 43% of the winning algorithms exceeded the state of the art but only 11% completely solved the respective domain problem. The insights of our study could help researchers (1) improve algorithm development strategies when approaching new problems, and (2) focus on open research questions revealed by this work.

Louise Bloch, Christoph M. Friedrich

Purpose: Hard-to-interpret Black-box Machine Learning (ML) were often used for early Alzheimer's Disease (AD) detection. Methods: To interpret eXtreme Gradient Boosting (XGBoost), Random Forest (RF), and Support Vector Machine (SVM) black-box models a workflow based on Shapley values was developed. All models were trained on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and evaluated for an independent ADNI test set, as well as the external Australian Imaging and Lifestyle flagship study of Ageing (AIBL), and Open Access Series of Imaging Studies (OASIS) datasets. Shapley values were compared to intuitively interpretable Decision Trees (DTs), and Logistic Regression (LR), as well as natural and permutation feature importances. To avoid the reduction of the explanation validity caused by correlated features, forward selection and aspect consolidation were implemented. Results: Some black-box models outperformed DTs and LR. The forward-selected features correspond to brain areas previously associated with AD. Shapley values identified biologically plausible associations with moderate to strong correlations with feature importances. The most important RF features to predict AD conversion were the volume of the amygdalae, and a cognitive test score. Good cognitive test performances and large brain volumes decreased the AD risk. The models trained using cognitive test scores significantly outperformed brain volumetric models ($p<0.05$). Cognitive Normal (CN) vs. AD models were successfully transferred to external datasets. Conclusion: In comparison to previous work, improved performances for ADNI and AIBL were achieved for CN vs. Mild Cognitive Impairment (MCI) classification using brain volumes. The Shapley values and the feature importances showed moderate to strong correlations.

Ahmad Idrissi-Yaghir, Henning Schäfer, Nadja Bauer et al.

Understanding customer feedback is becoming a necessity for companies to identify problems and improve their products and services. Text classification and sentiment analysis can play a major role in analyzing this data by using a variety of machine and deep learning approaches. In this work, different transformer-based models are utilized to explore how efficient these models are when working with a German customer feedback dataset. In addition, these pre-trained models are further analyzed to determine if adapting them to a specific domain using unlabeled data can yield better results than off-the-shelf pre-trained models. To evaluate the models, two downstream tasks from the GermEval 2017 are considered. The experimental results show that transformer-based models can reach significant improvements compared to a fastText baseline and outperform the published scores and previous models. For the subtask Relevance Classification, the best models achieve a micro-averaged $F1$-Score of 96.1 % on the first test set and 95.9 % on the second one, and a score of 85.1 % and 85.3 % for the subtask Polarity Classification.

Sven Koitka, Giulia Baldini, Cynthia S. Schmidt et al.

Traditional segmentation networks approach anatomical structures as standalone elements, overlooking the intrinsic hierarchical connections among them. This study introduces Softmax for Arbitrary Label Trees (SALT), a novel approach designed to leverage the hierarchical relationships between labels, improving the efficiency and interpretability of the segmentations. This study introduces a novel segmentation technique for CT imaging, which leverages conditional probabilities to map the hierarchical structure of anatomical landmarks, such as the spine's division into lumbar, thoracic, and cervical regions and further into individual vertebrae. The model was developed using the SAROS dataset from The Cancer Imaging Archive (TCIA), comprising 900 body region segmentations from 883 patients. The dataset was further enhanced by generating additional segmentations with the TotalSegmentator, for a total of 113 labels. The model was trained on 600 scans, while validation and testing were conducted on 150 CT scans. Performance was assessed using the Dice score across various datasets, including SAROS, CT-ORG, FLARE22, LCTSC, LUNA16, and WORD. Among the evaluated datasets, SALT achieved its best results on the LUNA16 and SAROS datasets, with Dice scores of 0.93 and 0.929 respectively. The model demonstrated reliable accuracy across other datasets, scoring 0.891 on CT-ORG and 0.849 on FLARE22. The LCTSC dataset showed a score of 0.908 and the WORD dataset also showed good performance with a score of 0.844. SALT used the hierarchical structures inherent in the human body to achieve whole-body segmentations with an average of 35 seconds for 100 slices. This rapid processing underscores its potential for integration into clinical workflows, facilitating the automatic and efficient computation of full-body segmentations with each CT scan, thus enhancing diagnostic processes and patient care.

Bohao Chu, Qianli Wang, Hendrik Damm et al.

How can system-generated responses be efficiently verified, especially in the high-stakes biomedical domain? To address this challenge, we introduce eTracer, a plug-and-play framework that enables traceable text generation by grounding claims against contextual evidence. Through post-hoc grounding, each response claim is aligned with contextual evidence that either supports or contradicts it. Building on claim-level grounding results, eTracer not only enables users to precisely trace responses back to their contextual source but also quantifies response faithfulness, thereby enabling the verifiability and trustworthiness of generated responses. Experiments show that our claim-level grounding approach alleviates the limitations of conventional grounding methods in aligning generated statements with contextual sentence-level evidence, resulting in substantial improvements in overall grounding quality and user verification efficiency. The code and data are available at https://github.com/chubohao/eTracer.

Johannes Rückert, Louise Bloch, Christoph M. Friedrich

Diagrams are widely used to visualize data in publications. The research field of data visualization deals with defining principles and guidelines for the creation and use of these diagrams, which are often not known or adhered to by researchers, leading to misinformation caused by providing inaccurate or incomplete information. In this work, large Vision Language Models (VLMs) are used to analyze diagrams in order to identify potential problems in regards to selected data visualization principles and guidelines. To determine the suitability of VLMs for these tasks, five open source VLMs and five prompting strategies are compared using a set of questions derived from selected data visualization guidelines. The results show that the employed VLMs work well to accurately analyze diagram types (F1-score 82.49 %), 3D effects (F1-score 98.55 %), axes labels (F1-score 76.74 %), lines (RMSE 1.16), colors (RMSE 1.60) and legends (F1-score 96.64 %, RMSE 0.70), while they cannot reliably provide feedback about the image quality (F1-score 0.74 %) and tick marks/labels (F1-score 46.13 %). Among the employed VLMs, Qwen2.5VL performs best, and the summarizing prompting strategy performs best for most of the experimental questions. It is shown that VLMs can be used to automatically identify a number of potential issues in diagrams, such as missing axes labels, missing legends, and unnecessary 3D effects. The approach laid out in this work can be extended for further aspects of data visualization.

Henning Schäfer, Cynthia S. Schmidt, Johannes Wutzkowsky et al.

Despite the growing adoption of electronic health records, many processes still rely on paper documents, reflecting the heterogeneous real-world conditions in which healthcare is delivered. The manual transcription process is time-consuming and prone to errors when transferring paper-based data to digital formats. To streamline this workflow, this study presents an open-source pipeline that extracts and categorizes checkbox data from scanned documents. Demonstrated on transfusion reaction reports, the design supports adaptation to other checkbox-rich document types. The proposed method integrates checkbox detection, multilingual optical character recognition (OCR) and multilingual vision-language models (VLMs). The pipeline achieves high precision and recall compared against annually compiled gold-standards from 2017 to 2024. The result is a reduction in administrative workload and accurate regulatory reporting. The open-source availability of this pipeline encourages self-hosted parsing of checkbox forms.

Johannes Rückert, Louise Bloch, Raphael Brüngel et al.

Automated medical image analysis systems often require large amounts of training data with high quality labels, which are difficult and time consuming to generate. This paper introduces Radiology Object in COntext version 2 (ROCOv2), a multimodal dataset consisting of radiological images and associated medical concepts and captions extracted from the PMC Open Access subset. It is an updated version of the ROCO dataset published in 2018, and adds 35,705 new images added to PMC since 2018. It further provides manually curated concepts for imaging modalities with additional anatomical and directional concepts for X-rays. The dataset consists of 79,789 images and has been used, with minor modifications, in the concept detection and caption prediction tasks of ImageCLEFmedical Caption 2023. The dataset is suitable for training image annotation models based on image-caption pairs, or for multi-label image classification using Unified Medical Language System (UMLS) concepts provided with each image. In addition, it can serve for pre-training of medical domain models, and evaluation of deep learning models for multi-task learning.

Hendrik Damm, Tabea M. G. Pakull, Bahadır Eryılmaz et al.

This study aims to leverage state of the art language models to automate generating the "Brief Hospital Course" and "Discharge Instructions" sections of Discharge Summaries from the MIMIC-IV dataset, reducing clinicians' administrative workload. We investigate how automation can improve documentation accuracy, alleviate clinician burnout, and enhance operational efficacy in healthcare facilities. This research was conducted within our participation in the Shared Task Discharge Me! at BioNLP @ ACL 2024. Various strategies were employed, including few-shot learning, instruction tuning, and Dynamic Expert Selection (DES), to develop models capable of generating the required text sections. Notably, utilizing an additional clinical domain-specific dataset demonstrated substantial potential to enhance clinical language processing. The DES method, which optimizes the selection of text outputs from multiple predictions, proved to be especially effective. It achieved the highest overall score of 0.332 in the competition, surpassing single-model outputs. This finding suggests that advanced deep learning methods in combination with DES can effectively automate parts of electronic health record documentation. These advancements could enhance patient care by freeing clinician time for patient interactions. The integration of text selection strategies represents a promising avenue for further research.

Ahmad Idrissi-Yaghir, Amin Dada, Henning Schäfer et al.

Recent advances in natural language processing (NLP) can be largely attributed to the advent of pre-trained language models such as BERT and RoBERTa. While these models demonstrate remarkable performance on general datasets, they can struggle in specialized domains such as medicine, where unique domain-specific terminologies, domain-specific abbreviations, and varying document structures are common. This paper explores strategies for adapting these models to domain-specific requirements, primarily through continuous pre-training on domain-specific data. We pre-trained several German medical language models on 2.4B tokens derived from translated public English medical data and 3B tokens of German clinical data. The resulting models were evaluated on various German downstream tasks, including named entity recognition (NER), multi-label classification, and extractive question answering. Our results suggest that models augmented by clinical and translation-based pre-training typically outperform general domain models in medical contexts. We conclude that continuous pre-training has demonstrated the ability to match or even exceed the performance of clinical models trained from scratch. Furthermore, pre-training on clinical data or leveraging translated texts have proven to be reliable methods for domain adaptation in medical NLP tasks.

Tabea M. G. Pakull, Hendrik Damm, Ahmad Idrissi-Yaghir et al.

This paper details the efforts of the WisPerMed team in the BioLaySumm2024 Shared Task on automatic lay summarization in the biomedical domain, aimed at making scientific publications accessible to non-specialists. Large language models (LLMs), specifically the BioMistral and Llama3 models, were fine-tuned and employed to create lay summaries from complex scientific texts. The summarization performance was enhanced through various approaches, including instruction tuning, few-shot learning, and prompt variations tailored to incorporate specific context information. The experiments demonstrated that fine-tuning generally led to the best performance across most evaluated metrics. Few-shot learning notably improved the models' ability to generate relevant and factually accurate texts, particularly when using a well-crafted prompt. Additionally, a Dynamic Expert Selection (DES) mechanism to optimize the selection of text outputs based on readability and factuality metrics was developed. Out of 54 participants, the WisPerMed team reached the 4th place, measured by readability, factuality, and relevance. Determined by the overall score, our approach improved upon the baseline by approx. 5.5 percentage points and was only approx 1.5 percentage points behind the first place.

Martin Sondermann, Pinar Bisgin, Niklas Tschorn et al.

The automated classification of phonocardiogram (PCG) recordings represents a substantial advancement in cardiovascular diagnostics. This paper presents a systematic comparison of four distinct models for heart murmur detection: two specialized convolutional neural networks (CNNs) and two zero-shot universal audio transformers (BEATs), evaluated using fixed-length and heart cycle normalization approaches. Utilizing the PhysioNet2022 dataset, a custom heart cycle normalization method tailored to individual cardiac rhythms is introduced. The findings indicate the following AUROC values: the CNN model with fixed-length windowing achieves 79.5%, the CNN model with heart cycle normalization scores 75.4%, the BEATs transformer with fixed-length windowing achieves 65.7%, and the BEATs transformer with heart cycle normalization results in 70.1%. The findings indicate that physiological signal constraints, especially those introduced by different normalization strategies, have a substantial impact on model performance. The research provides evidence-based guidelines for architecture selection in clinical settings, emphasizing the need for a balance between accuracy and computational efficiency. Although specialized CNNs demonstrate superior performance overall, the zero-shot transformer models may offer promising efficiency advantages during development, such as faster training and evaluation cycles, despite their lower classification accuracy. These findings highlight the potential of automated classification systems to enhance cardiac diagnostics and improve patient care.

Louise Bloch, Raphael Brüngel, Christoph M. Friedrich

Diabetic foot ulcers are a common manifestation of lesions on the diabetic foot, a syndrome acquired as a long-term complication of diabetes mellitus. Accompanying neuropathy and vascular damage promote acquisition of pressure injuries and tissue death due to ischaemia. Affected areas are prone to infections, hindering the healing progress. The research at hand investigates an approach on classification of infection and ischaemia, conducted as part of the Diabetic Foot Ulcer Challenge (DFUC) 2021. Different models of the EfficientNet family are utilized in ensembles. An extension strategy for the training data is applied, involving pseudo-labeling for unlabeled images, and extensive generation of synthetic images via pix2pixHD to cope with severe class imbalances. The resulting extended training dataset features $8.68$ times the size of the baseline and shows a real to synthetic image ratio of $1:3$. Performances of models and ensembles trained on the baseline and extended training dataset are compared. Synthetic images featured a broad qualitative variety. Results show that models trained on the extended training dataset as well as their ensemble benefit from the large extension. F1-Scores for rare classes receive outstanding boosts, while those for common classes are either not harmed or boosted moderately. A critical discussion concretizes benefits and identifies limitations, suggesting improvements. The work concludes that classification performance of individual models as well as that of ensembles can be boosted utilizing synthetic images. Especially performance for rare classes benefits notably.

Moi Hoon Yap, Ryo Hachiuma, Azadeh Alavi et al.

There has been a substantial amount of research involving computer methods and technology for the detection and recognition of diabetic foot ulcers (DFUs), but there is a lack of systematic comparisons of state-of-the-art deep learning object detection frameworks applied to this problem. DFUC2020 provided participants with a comprehensive dataset consisting of 2,000 images for training and 2,000 images for testing. This paper summarises the results of DFUC2020 by comparing the deep learning-based algorithms proposed by the winning teams: Faster R-CNN, three variants of Faster R-CNN and an ensemble method; YOLOv3; YOLOv5; EfficientDet; and a new Cascade Attention Network. For each deep learning method, we provide a detailed description of model architecture, parameter settings for training and additional stages including pre-processing, data augmentation and post-processing. We provide a comprehensive evaluation for each method. All the methods required a data augmentation stage to increase the number of images available for training and a post-processing stage to remove false positives. The best performance was obtained from Deformable Convolution, a variant of Faster R-CNN, with a mean average precision (mAP) of 0.6940 and an F1-Score of 0.7434. Finally, we demonstrate that the ensemble method based on different deep learning methods can enhanced the F1-Score but not the mAP.

Marcel Trotzek, Sven Koitka, Christoph M. Friedrich

Depression is ranked as the largest contributor to global disability and is also a major reason for suicide. Still, many individuals suffering from forms of depression are not treated for various reasons. Previous studies have shown that depression also has an effect on language usage and that many depressed individuals use social media platforms or the internet in general to get information or discuss their problems. This paper addresses the early detection of depression using machine learning models based on messages on a social platform. In particular, a convolutional neural network based on different word embeddings is evaluated and compared to a classification based on user-level linguistic metadata. An ensemble of both approaches is shown to achieve state-of-the-art results in a current early detection task. Furthermore, the currently popular ERDE score as metric for early detection systems is examined in detail and its drawbacks in the context of shared tasks are illustrated. A slightly modified metric is proposed and compared to the original score. Finally, a new word embedding was trained on a large corpus of the same domain as the described task and is evaluated as well.