Marcus R. Makowski

Tobias Czempiel, Coco Rogers, Matthias Keicher et al. · stanford

Quantifying COVID-19 infection over time is an important task to manage the hospitalization of patients during a global pandemic. Recently, deep learning-based approaches have been proposed to help radiologists automatically quantify COVID-19 pathologies on longitudinal CT scans. However, the learning process of deep learning methods demands extensive training data to learn the complex characteristics of infected regions over longitudinal scans. It is challenging to collect a large-scale dataset, especially for longitudinal training. In this study, we want to address this problem by proposing a new self-supervised learning method to effectively train longitudinal networks for the quantification of COVID-19 infections. For this purpose, longitudinal self-supervision schemes are explored on clinical longitudinal COVID-19 CT scans. Experimental results show that the proposed method is effective, helping the model better exploit the semantics of longitudinal data and improve two COVID-19 quantification tasks.

Lisa C. Adams, Felix Busch, Daniel Truhn et al.

Generative models such as DALL-E 2 could represent a promising future tool for image generation, augmentation, and manipulation for artificial intelligence research in radiology provided that these models have sufficient medical domain knowledge. Here we show that DALL-E 2 has learned relevant representations of X-ray images with promising capabilities in terms of zero-shot text-to-image generation of new images, continuation of an image beyond its original boundaries, or removal of elements, while pathology generation or CT, MRI, and ultrasound images are still limited. The use of generative models for augmenting and generating radiological data thus seems feasible, even if further fine-tuning and adaptation of these models to the respective domain is required beforehand.

Keno K. Bressem, Jens-Michalis Papaioannou, Paul Grundmann et al.

This paper presents medBERTde, a pre-trained German BERT model specifically designed for the German medical domain. The model has been trained on a large corpus of 4.7 Million German medical documents and has been shown to achieve new state-of-the-art performance on eight different medical benchmarks covering a wide range of disciplines and medical document types. In addition to evaluating the overall performance of the model, this paper also conducts a more in-depth analysis of its capabilities. We investigate the impact of data deduplication on the model's performance, as well as the potential benefits of using more efficient tokenization methods. Our results indicate that domain-specific models such as medBERTde are particularly useful for longer texts, and that deduplication of training data does not necessarily lead to improved performance. Furthermore, we found that efficient tokenization plays only a minor role in improving model performance, and attribute most of the improved performance to the large amount of training data. To encourage further research, the pre-trained model weights and new benchmarks based on radiological data are made publicly available for use by the scientific community.

Soroosh Tayebi Arasteh, Alexander Ziller, Christiane Kuhl et al.

Artificial intelligence (AI) models are increasingly used in the medical domain. However, as medical data is highly sensitive, special precautions to ensure its protection are required. The gold standard for privacy preservation is the introduction of differential privacy (DP) to model training. Prior work indicates that DP has negative implications on model accuracy and fairness, which are unacceptable in medicine and represent a main barrier to the widespread use of privacy-preserving techniques. In this work, we evaluated the effect of privacy-preserving training of AI models regarding accuracy and fairness compared to non-private training. For this, we used two datasets: (1) A large dataset (N=193,311) of high quality clinical chest radiographs, and (2) a dataset (N=1,625) of 3D abdominal computed tomography (CT) images, with the task of classifying the presence of pancreatic ductal adenocarcinoma (PDAC). Both were retrospectively collected and manually labeled by experienced radiologists. We then compared non-private deep convolutional neural networks (CNNs) and privacy-preserving (DP) models with respect to privacy-utility trade-offs measured as area under the receiver-operator-characteristic curve (AUROC), and privacy-fairness trade-offs, measured as Pearson's r or Statistical Parity Difference. We found that, while the privacy-preserving trainings yielded lower accuracy, they did largely not amplify discrimination against age, sex or co-morbidity. Our study shows that -- under the challenging realistic circumstances of a real-life clinical dataset -- the privacy-preserving training of diagnostic deep learning models is possible with excellent diagnostic accuracy and fairness.

Felix J. Dorfner, Amin Dada, Felix Busch et al.

Large language models (LLMs) have shown potential in biomedical applications, leading to efforts to fine-tune them on domain-specific data. However, the effectiveness of this approach remains unclear. This study evaluates the performance of biomedically fine-tuned LLMs against their general-purpose counterparts on a variety of clinical tasks. We evaluated their performance on clinical case challenges from the New England Journal of Medicine (NEJM) and the Journal of the American Medical Association (JAMA) and on several clinical tasks (e.g., information extraction, document summarization, and clinical coding). Using benchmarks specifically chosen to be likely outside the fine-tuning datasets of biomedical models, we found that biomedical LLMs mostly perform inferior to their general-purpose counterparts, especially on tasks not focused on medical knowledge. While larger models showed similar performance on case tasks (e.g., OpenBioLLM-70B: 66.4% vs. Llama-3-70B-Instruct: 65% on JAMA cases), smaller biomedical models showed more pronounced underperformance (e.g., OpenBioLLM-8B: 30% vs. Llama-3-8B-Instruct: 64.3% on NEJM cases). Similar trends were observed across the CLUE (Clinical Language Understanding Evaluation) benchmark tasks, with general-purpose models often performing better on text generation, question answering, and coding tasks. Our results suggest that fine-tuning LLMs to biomedical data may not provide the expected benefits and may potentially lead to reduced performance, challenging prevailing assumptions about domain-specific adaptation of LLMs and highlighting the need for more rigorous evaluation frameworks in healthcare AI. Alternative approaches, such as retrieval-augmented generation, may be more effective in enhancing the biomedical capabilities of LLMs without compromising their general knowledge.

Felix Busch, Tianyu Han, Marcus Makowski et al.

The study evaluates and compares GPT-4 and GPT-4Vision for radiological tasks, suggesting GPT-4Vision may recognize radiological features from images, thereby enhancing its diagnostic potential over text-based descriptions.

Sebastian Ziegelmayer, Alexander W. Marka, Nicolas Lenhart et al.

The remarkable generative capabilities of multimodal foundation models are currently being explored for a variety of applications. Generating radiological impressions is a challenging task that could significantly reduce the workload of radiologists. In our study we explored and analyzed the generative abilities of GPT-4 for Chest X-ray impression generation. To generate and evaluate impressions of chest X-rays based on different input modalities (image, text, text and image), a blinded radiological report was written for 25-cases of the publicly available NIH-dataset. GPT-4 was given image, finding section or both sequentially to generate an input dependent impression. In a blind randomized reading, 4-radiologists rated the impressions and were asked to classify the impression origin (Human, AI), providing justification for their decision. Lastly text model evaluation metrics and their correlation with the radiological score (summation of the 4 dimensions) was assessed. According to the radiological score, the human-written impression was rated highest, although not significantly different to text-based impressions. The automated evaluation metrics showed moderate to substantial correlations to the radiological score for the image impressions, however individual scores were highly divergent among inputs, indicating insufficient representation of radiological quality. Detection of AI-generated impressions varied by input and was 61% for text-based impressions. Impressions classified as AI-generated had significantly worse radiological scores even when written by a radiologist, indicating potential bias. Our study revealed significant discrepancies between a radiological assessment and common automatic evaluation metrics depending on the model input. The detection of AI-generated findings is subject to bias that highly rated impressions are perceived as human-written.

Felix J. Dorfner, Janis L. Vahldiek, Leonhard Donle et al.

Purpose: To examine whether incorporating anatomical awareness into a deep learning model can improve generalizability and enable prediction of disease progression. Methods: This retrospective multicenter study included conventional pelvic radiographs of 4 different patient cohorts focusing on axial spondyloarthritis (axSpA) collected at university and community hospitals. The first cohort, which consisted of 1483 radiographs, was split into training (n=1261) and validation (n=222) sets. The other cohorts comprising 436, 340, and 163 patients, respectively, were used as independent test datasets. For the second cohort, follow-up data of 311 patients was used to examine progression prediction capabilities. Two neural networks were trained, one on images cropped to the bounding box of the sacroiliac joints (anatomy-aware) and the other one on full radiographs. The performance of the models was compared using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Results: On the three test datasets, the standard model achieved AUC scores of 0.853, 0.817, 0.947, with an accuracy of 0.770, 0.724, 0.850. Whereas the anatomy-aware model achieved AUC scores of 0.899, 0.846, 0.957, with an accuracy of 0.821, 0.744, 0.906, respectively. The patients who were identified as high risk by the anatomy aware model had an odds ratio of 2.16 (95% CI: 1.19, 3.86) for having progression of radiographic sacroiliitis within 2 years. Conclusion: Anatomical awareness can improve the generalizability of a deep learning model in detecting radiographic sacroiliitis. The model is published as fully open source alongside this study.

Bastien Le Guellec, Kokou Adambounou, Lisa C Adams et al.

Rationale and Objectives: To develop and validate PARROT (Polyglottal Annotated Radiology Reports for Open Testing), a large, multicentric, open-access dataset of fictional radiology reports spanning multiple languages for testing natural language processing applications in radiology. Materials and Methods: From May to September 2024, radiologists were invited to contribute fictional radiology reports following their standard reporting practices. Contributors provided at least 20 reports with associated metadata including anatomical region, imaging modality, clinical context, and for non-English reports, English translations. All reports were assigned ICD-10 codes. A human vs. AI report differentiation study was conducted with 154 participants (radiologists, healthcare professionals, and non-healthcare professionals) assessing whether reports were human-authored or AI-generated. Results: The dataset comprises 2,658 radiology reports from 76 authors across 21 countries and 13 languages. Reports cover multiple imaging modalities (CT: 36.1%, MRI: 22.8%, radiography: 19.0%, ultrasound: 16.8%) and anatomical regions, with chest (19.9%), abdomen (18.6%), head (17.3%), and pelvis (14.1%) being most prevalent. In the differentiation study, participants achieved 53.9% accuracy (95% CI: 50.7%-57.1%) in distinguishing between human and AI-generated reports, with radiologists performing significantly better (56.9%, 95% CI: 53.3%-60.6%, p<0.05) than other groups. Conclusion: PARROT represents the largest open multilingual radiology report dataset, enabling development and validation of natural language processing applications across linguistic, geographic, and clinical boundaries without privacy constraints.

Michelle Xiao-Lin Foo, Seong Tae Kim, Magdalini Paschali et al.

Consistent segmentation of COVID-19 patient's CT scans across multiple time points is essential to assess disease progression and response to therapy accurately. Existing automatic and interactive segmentation models for medical images only use data from a single time point (static). However, valuable segmentation information from previous time points is often not used to aid the segmentation of a patient's follow-up scans. Also, fully automatic segmentation techniques frequently produce results that would need further editing for clinical use. In this work, we propose a new single network model for interactive segmentation that fully utilizes all available past information to refine the segmentation of follow-up scans. In the first segmentation round, our model takes 3D volumes of medical images from two-time points (target and reference) as concatenated slices with the additional reference time point segmentation as a guide to segment the target scan. In subsequent segmentation refinement rounds, user feedback in the form of scribbles that correct the segmentation and the target's previous segmentation results are additionally fed into the model. This ensures that the segmentation information from previous refinement rounds is retained. Experimental results on our in-house multiclass longitudinal COVID-19 dataset show that the proposed model outperforms its static version and can assist in localizing COVID-19 infections in patient's follow-up scans.

Markus Krönke, Christine Eilers, Desislava Dimova et al.

Background: Thyroid volumetry is crucial in diagnosis, treatment and monitoring of thyroid diseases. However, conventional thyroid volumetry with 2D ultrasound is highly operator-dependent. This study compares 2D ultrasound and tracked 3D ultrasound with an automatic thyroid segmentation based on a deep neural network regarding inter- and intraobserver variability, time and accuracy. Volume reference was MRI. Methods: 28 healthy volunteers were scanned with 2D and 3D ultrasound as well as by MRI. Three physicians (MD 1, 2, 3) with different levels of experience (6, 4 and 1 a) performed three 2D ultrasound and three tracked 3D ultrasound scans on each volunteer. In the 2D scans the thyroid lobe volumes were calculated with the ellipsoid formula. A convolutional deep neural network (CNN) segmented the 3D thyroid lobes automatically. On MRI (T1 VIBE sequence) the thyroid was manually segmented by an experienced medical doctor. Results: The CNN was trained to obtain a dice score of 0.94. The interobserver variability comparing two MDs showed mean differences for 2D and 3D respectively of 0.58 ml to 0.52 ml (MD1 vs. 2), -1.33 ml to -0.17 ml (MD1 vs. 3) and -1.89 ml to -0.70 ml (MD2 vs. 3). Paired samples t-tests showed significant differences in two comparisons for 2D and none for 3D. Intraobsever variability was similar for 2D and 3D ultrasound. Comparison of ultrasound volumes and MRI volumes by paired samples t-tests showed a significant difference for the 2D volumetry of all MDs, and no significant difference for 3D ultrasound. Acquisition time was significantly shorter for 3D ultrasound. Conclusion: Tracked 3D ultrasound combined with a CNN segmentation significantly reduces interobserver variability in thyroid volumetry and increases the accuracy of the measurements with shorter acquisition times.

Moritz Knolle, Dmitrii Usynin, Alexander Ziller et al.

The application of differential privacy to the training of deep neural networks holds the promise of allowing large-scale (decentralized) use of sensitive data while providing rigorous privacy guarantees to the individual. The predominant approach to differentially private training of neural networks is DP-SGD, which relies on norm-based gradient clipping as a method for bounding sensitivity, followed by the addition of appropriately calibrated Gaussian noise. In this work we propose NeuralDP, a technique for privatising activations of some layer within a neural network, which by the post-processing properties of differential privacy yields a differentially private network. We experimentally demonstrate on two datasets (MNIST and Pediatric Pneumonia Dataset (PPD)) that our method offers substantially improved privacy-utility trade-offs compared to DP-SGD.

Matthias Keicher, Hendrik Burwinkel, David Bani-Harouni et al.

During the first wave of COVID-19, hospitals were overwhelmed with the high number of admitted patients. An accurate prediction of the most likely individual disease progression can improve the planning of limited resources and finding the optimal treatment for patients. However, when dealing with a newly emerging disease such as COVID-19, the impact of patient- and disease-specific factors (e.g. body weight or known co-morbidities) on the immediate course of disease is by and large unknown. In the case of COVID-19, the need for intensive care unit (ICU) admission of pneumonia patients is often determined only by acute indicators such as vital signs (e.g. breathing rate, blood oxygen levels), whereas statistical analysis and decision support systems that integrate all of the available data could enable an earlier prognosis. To this end, we propose a holistic graph-based approach combining both imaging and non-imaging information. Specifically, we introduce a multimodal similarity metric to build a population graph for clustering patients and an image-based end-to-end Graph Attention Network to process this graph and predict the COVID-19 patient outcomes: admission to ICU, need for ventilation and mortality. Additionally, the network segments chest CT images as an auxiliary task and extracts image features and radiomics for feature fusion with the available metadata. Results on a dataset collected in Klinikum rechts der Isar in Munich, Germany show that our approach outperforms single modality and non-graph baselines. Moreover, our clustering and graph attention allow for increased understanding of the patient relationships within the population graph and provide insight into the network's decision-making process.

Moritz Knolle, Alexander Ziller, Dmitrii Usynin et al.

We show that differentially private stochastic gradient descent (DP-SGD) can yield poorly calibrated, overconfident deep learning models. This represents a serious issue for safety-critical applications, e.g. in medical diagnosis. We highlight and exploit parallels between stochastic gradient Langevin dynamics, a scalable Bayesian inference technique for training deep neural networks, and DP-SGD, in order to train differentially private, Bayesian neural networks with minor adjustments to the original (DP-SGD) algorithm. Our approach provides considerably more reliable uncertainty estimates than DP-SGD, as demonstrated empirically by a reduction in expected calibration error (MNIST $\sim{5}$-fold, Pediatric Pneumonia Dataset $\sim{2}$-fold).

Alexander Ziller, Dmitrii Usynin, Moritz Knolle et al.

In recent years, formal methods of privacy protection such as differential privacy (DP), capable of deployment to data-driven tasks such as machine learning (ML), have emerged. Reconciling large-scale ML with the closed-form reasoning required for the principled analysis of individual privacy loss requires the introduction of new tools for automatic sensitivity analysis and for tracking an individual's data and their features through the flow of computation. For this purpose, we introduce a novel \textit{hybrid} automatic differentiation (AD) system which combines the efficiency of reverse-mode AD with an ability to obtain a closed-form expression for any given quantity in the computational graph. This enables modelling the sensitivity of arbitrary differentiable function compositions, such as the training of neural networks on private data. We demonstrate our approach by analysing the individual DP guarantees of statistical database queries. Moreover, we investigate the application of our technique to the training of DP neural networks. Our approach can enable the principled reasoning about privacy loss in the setting of data processing, and further the development of automatic sensitivity analysis and privacy budgeting systems.

Alexander Ziller, Dmitrii Usynin, Nicolas Remerscheid et al.

Collaborative machine learning techniques such as federated learning (FL) enable the training of models on effectively larger datasets without data transfer. Recent initiatives have demonstrated that segmentation models trained with FL can achieve performance similar to locally trained models. However, FL is not a fully privacy-preserving technique and privacy-centred attacks can disclose confidential patient data. Thus, supplementing FL with privacy-enhancing technologies (PTs) such as differential privacy (DP) is a requirement for clinical applications in a multi-institutional setting. The application of PTs to FL in medical imaging and the trade-offs between privacy guarantees and model utility, the ramifications on training performance and the susceptibility of the final models to attacks have not yet been conclusively investigated. Here we demonstrate the first application of differentially private gradient descent-based FL on the task of semantic segmentation in computed tomography. We find that high segmentation performance is possible under strong privacy guarantees with an acceptable training time penalty. We furthermore demonstrate the first successful gradient-based model inversion attack on a semantic segmentation model and show that the application of DP prevents it from divulging sensitive image features.

Keno K. Bressem, Stefan M. Niehues, Bernd Hamm et al.

Segmentation of pulmonary infiltrates can help assess severity of COVID-19, but manual segmentation is labor and time-intensive. Using neural networks to segment pulmonary infiltrates would enable automation of this task. However, training a 3D U-Net from computed tomography (CT) data is time- and resource-intensive. In this work, we therefore developed and tested a solution on how transfer learning can be used to train state-of-the-art segmentation models on limited hardware and in shorter time. We use the recently published RSNA International COVID-19 Open Radiology Database (RICORD) to train a fully three-dimensional U-Net architecture using an 18-layer 3D ResNet, pretrained on the Kinetics-400 dataset as encoder. The generalization of the model was then tested on two openly available datasets of patients with COVID-19, who received chest CTs (Corona Cases and MosMed datasets). Our model performed comparable to previously published 3D U-Net architectures, achieving a mean Dice score of 0.679 on the tuning dataset, 0.648 on the Coronacases dataset and 0.405 on the MosMed dataset. Notably, these results were achieved with shorter training time on a single GPU with less memory available than the GPUs used in previous studies.

Alexander Ziller, Jonathan Passerat-Palmbach, Théo Ryffel et al.

The utilisation of artificial intelligence in medicine and healthcare has led to successful clinical applications in several domains. The conflict between data usage and privacy protection requirements in such systems must be resolved for optimal results as well as ethical and legal compliance. This calls for innovative solutions such as privacy-preserving machine learning (PPML). We present PriMIA (Privacy-preserving Medical Image Analysis), a software framework designed for PPML in medical imaging. In a real-life case study we demonstrate significantly better classification performance of a securely aggregated federated learning model compared to human experts on unseen datasets. Furthermore, we show an inference-as-a-service scenario for end-to-end encrypted diagnosis, where neither the data nor the model are revealed. Lastly, we empirically evaluate the framework's security against a gradient-based model inversion attack and demonstrate that no usable information can be recovered from the model.

Moritz Knolle, Georgios Kaissis, Friederike Jungmann et al.

For artificial intelligence-based image analysis methods to reach clinical applicability, the development of high-performance algorithms is crucial. For example, existent segmentation algorithms based on natural images are neither efficient in their parameter use nor optimized for medical imaging. Here we present MoNet, a highly optimized neural-network-based pancreatic segmentation algorithm focused on achieving high performance by efficient multi-scale image feature utilization.