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.
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
The growing impact of preprint servers enables the rapid sharing of time-sensitive research. Likewise, it is becoming increasingly difficult to distinguish high-quality, peer-reviewed research from preprints. Although preprints are often later published in peer-reviewed journals, this information is often missing from preprint servers. To overcome this problem, the PreprintResolver was developed, which uses four literature databases (DBLP, SemanticScholar, OpenAlex, and CrossRef / CrossCite) to identify preprint-publication pairs for the arXiv preprint server. The target audience focuses on, but is not limited to inexperienced researchers and students, especially from the field of computer science. The tool is based on a fuzzy matching of author surnames, titles, and DOIs. Experiments were performed on a sample of 1,000 arXiv-preprints from the research field of computer science and without any publication information. With 77.94 %, computer science is highly affected by missing publication information in arXiv. The results show that the PreprintResolver was able to resolve 603 out of 1,000 (60.3 %) arXiv-preprints from the research field of computer science and without any publication information. All four literature databases contributed to the final result. In a manual validation, a random sample of 100 resolved preprints was checked. For all preprints, at least one result is plausible. For nine preprints, more than one result was identified, three of which are partially invalid. In conclusion the PreprintResolver is suitable for individual, manually reviewed requests, but less suitable for bulk requests. The PreprintResolver tool (https://preprintresolver.eu, Available from 2023-08-01) and source code (https://gitlab.com/ippolis_wp3/preprint-resolver, Accessed: 2023-07-19) is available online.
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.
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.
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.
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.
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.
There has been a substantial amount of research on 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. With recent development and data sharing performed as part of the DFU Challenge (DFUC2020) such a comparison becomes possible: DFUC2020 provided participants with a comprehensive dataset consisting of 2,000 images for training each method and 2,000 images for testing them. The following deep learning-based algorithms are compared in this paper: 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 is obtained Deformable Convolution, a variant of Faster R-CNN, with a 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. Our results show that state-of-the-art deep learning methods can detect DFU with some accuracy, but there are many challenges ahead before they can be implemented in real world settings.