Rainer Fietkau

Thomas Weissmann, Yixing Huang, Stefan Fischer et al.

Background: Deep learning (DL)-based head and neck lymph node level (HN_LNL) autodelineation is of high relevance to radiotherapy research and clinical treatment planning but still underinvestigated in academic literature. Methods: An expert-delineated cohort of 35 planning CTs was used for training of an nnU-net 3D-fullres/2D-ensemble model for autosegmentation of 20 different HN_LNL. A second cohort acquired at the same institution later in time served as the test set (n=20). In a completely blinded evaluation, 3 clinical experts rated the quality of DL autosegmentations in a head-to-head comparison with expert-created contours. For a subgroup of 10 cases, intraobserver variability was compared to the average DL autosegmentation accuracy on the original and recontoured set of expert segmentations. A postprocessing step to adjust craniocaudal boundaries of level autosegmentations to the CT slice plane was introduced and the effect on geometric accuracy and expert rating was investigated. Results: Blinded expert ratings for DL segmentations and expert-created contours were not significantly different. DL segmentations with slice plane adjustment were rated numerically higher (mean, 81.0 vs. 79.6,p=0.185) and DL segmentations without slice plane adjustment were rated numerically lower (77.2 vs. 79.6,p=0.167) than manually drawn contours. DL segmentations with CT slice plane adjustment were rated significantly better than DL contours without slice plane adjustment (81.0 vs. 77.2,p=0.004). Geometric accuracy of DL segmentations was not different from intraobserver variability (mean, 0.76 vs. 0.77, p=0.307). Conclusions: We show that a nnU-net 3D-fullres/2D-ensemble model can be used for highly accurate autodelineation of HN_LNL using only a limited training dataset that is ideally suited for large-scale standardized autodelineation of HN_LNL in the research setting.

Yixing Huang, Christoph Bert, Stefan Fischer et al.

Due to data privacy constraints, data sharing among multiple centers is restricted. Continual learning, as one approach to peer-to-peer federated learning, can promote multicenter collaboration on deep learning algorithm development by sharing intermediate models instead of training data. This work aims to investigate the feasibility of continual learning for multicenter collaboration on an exemplary application of brain metastasis identification using DeepMedic. 920 T1 MRI contrast enhanced volumes are split to simulate multicenter collaboration scenarios. A continual learning algorithm, synaptic intelligence (SI), is applied to preserve important model weights for training one center after another. In a bilateral collaboration scenario, continual learning with SI achieves a sensitivity of 0.917, and naive continual learning without SI achieves a sensitivity of 0.906, while two models trained on internal data solely without continual learning achieve sensitivity of 0.853 and 0.831 only. In a seven-center multilateral collaboration scenario, the models trained on internal datasets (100 volumes each center) without continual learning obtain a mean sensitivity value of 0.699. With single-visit continual learning (i.e., the shared model visits each center only once during training), the sensitivity is improved to 0.788 and 0.849 without SI and with SI, respectively. With iterative continual learning (i.e., the shared model revisits each center multiple times during training), the sensitivity is further improved to 0.914, which is identical to the sensitivity using mixed data for training. Our experiments demonstrate that continual learning can improve brain metastasis identification performance for centers with limited data. This study demonstrates the feasibility of applying continual learning for peer-to-peer federated learning in multicenter collaboration.

Florian Putz, Johanna Grigo, Thomas Weissmann et al.

Background: Tumor segmentation in MRI is crucial in radiotherapy (RT) treatment planning for brain tumor patients. Segment anything (SA), a novel promptable foundation model for autosegmentation, has shown high accuracy for multiple segmentation tasks but was not evaluated on medical datasets yet. Methods: SA was evaluated in a point-to-mask task for glioma brain tumor autosegmentation on 16744 transversal slices from 369 MRI datasets (BraTS 2020). Up to 9 point prompts were placed per slice. Tumor core (enhancing tumor + necrotic core) was segmented on contrast-enhanced T1w sequences. Out of the 3 masks predicted by SA, accuracy was evaluated for the mask with the highest calculated IoU (oracle mask) and with highest model predicted IoU (suggested mask). In addition to assessing SA on whole MRI slices, SA was also evaluated on images cropped to the tumor (max. 3D extent + 2 cm). Results: Mean best IoU (mbIoU) using oracle mask on full MRI slices was 0.762 (IQR 0.713-0.917). Best 2D mask was achieved after a mean of 6.6 point prompts (IQR 5-9). Segmentation accuracy was significantly better for high- compared to low-grade glioma cases (mbIoU 0.789 vs. 0.668). Accuracy was worse using MRI slices cropped to the tumor (mbIoU 0.759) and was much worse using suggested mask (full slices 0.572). For all experiments, accuracy was low on peripheral slices with few tumor voxels (mbIoU, <300: 0.537 vs. >=300: 0.841). Stacking best oracle segmentations from full axial MRI slices, mean 3D DSC for tumor core was 0.872, which was improved to 0.919 by combining axial, sagittal and coronal masks. Conclusions: The Segment Anything foundation model, while trained on photos, can achieve high zero-shot accuracy for glioma brain tumor segmentation on MRI slices. The results suggest that Segment Anything can accelerate and facilitate RT treatment planning, when properly integrated in a clinical application.

Yixing Huang, Ahmed Gomaa, Sabine Semrau et al.

The potential of large language models in medicine for education and decision making purposes has been demonstrated as they achieve decent scores on medical exams such as the United States Medical Licensing Exam (USMLE) and the MedQA exam. In this work, we evaluate the performance of ChatGPT-4 in the specialized field of radiation oncology using the 38th American College of Radiology (ACR) radiation oncology in-training (TXIT) exam and the 2022 Red Journal Gray Zone cases. For the TXIT exam, ChatGPT-3.5 and ChatGPT-4 have achieved the scores of 63.65% and 74.57%, respectively, highlighting the advantage of the latest ChatGPT-4 model. Based on the TXIT exam, ChatGPT-4's strong and weak areas in radiation oncology are identified to some extent. Specifically, ChatGPT-4 demonstrates better knowledge of statistics, CNS & eye, pediatrics, biology, and physics than knowledge of bone & soft tissue and gynecology, as per the ACR knowledge domain. Regarding clinical care paths, ChatGPT-4 performs better in diagnosis, prognosis, and toxicity than brachytherapy and dosimetry. It lacks proficiency in in-depth details of clinical trials. For the Gray Zone cases, ChatGPT-4 is able to suggest a personalized treatment approach to each case with high correctness and comprehensiveness. Importantly, it provides novel treatment aspects for many cases, which are not suggested by any human experts. Both evaluations demonstrate the potential of ChatGPT-4 in medical education for the general public and cancer patients, as well as the potential to aid clinical decision-making, while acknowledging its limitations in certain domains. Because of the risk of hallucination, facts provided by ChatGPT always need to be verified.

Yihao Hou, Christoph Bert, Ahmed Gomaa et al.

Generating physician letters is a time-consuming task in daily clinical practice. This study investigates local fine-tuning of large language models (LLMs), specifically LLaMA models, for physician letter generation in a privacy-preserving manner within the field of radiation oncology. Our findings demonstrate that base LLaMA models, without fine-tuning, are inadequate for effectively generating physician letters. The QLoRA algorithm provides an efficient method for local intra-institutional fine-tuning of LLMs with limited computational resources (i.e., a single 48 GB GPU workstation within the hospital). The fine-tuned LLM successfully learns radiation oncology-specific information and generates physician letters in an institution-specific style. ROUGE scores of the generated summary reports highlight the superiority of the 8B LLaMA-3 model over the 13B LLaMA-2 model. Further multidimensional physician evaluations of 10 cases reveal that, although the fine-tuned LLaMA-3 model has limited capacity to generate content beyond the provided input data, it successfully generates salutations, diagnoses and treatment histories, recommendations for further treatment, and planned schedules. Overall, clinical benefit was rated highly by the clinical experts (average score of 3.44 on a 4-point scale). With careful physician review and correction, automated LLM-based physician letter generation has significant practical value.

Amr Hagag, Ahmed Gomaa, Dominik Kornek et al.

Survival prediction for cancer patients is critical for optimal treatment selection and patient management. Current patient survival prediction methods typically extract survival information from patients' clinical record data or biological and imaging data. In practice, experienced clinicians can have a preliminary assessment of patients' health status based on patients' observable physical appearances, which are mainly facial features. However, such assessment is highly subjective. In this work, the efficacy of objectively capturing and using prognostic information contained in conventional portrait photographs using deep learning for survival predication purposes is investigated for the first time. A pre-trained StyleGAN2 model is fine-tuned on a custom dataset of our cancer patients' photos to empower its generator with generative ability suitable for patients' photos. The StyleGAN2 is then used to embed the photographs to its highly expressive latent space. Utilizing the state-of-the-art survival analysis models and based on StyleGAN's latent space photo embeddings, this approach achieved a C-index of 0.677, which is notably higher than chance and evidencing the prognostic value embedded in simple 2D facial images. In addition, thanks to StyleGAN's interpretable latent space, our survival prediction model can be validated for relying on essential facial features, eliminating any biases from extraneous information like clothing or background. Moreover, a health attribute is obtained from regression coefficients, which has important potential value for patient care.

Philipp Sommer, Yixing Huang, Christoph Bert et al.

Stereotactic radiotherapy (SRT) is one of the most important treatment for patients with brain metastases (BM). Conventionally, following SRT patients are monitored by serial imaging and receive salvage treatments in case of significant tumor growth. We hypothesized that using radiomics and machine learning (ML), metastases at high risk for subsequent progression could be identified during follow-up prior to the onset of significant tumor growth, enabling personalized follow-up intervals and early selection for salvage treatment. All experiments are performed on a dataset from clinical routine of the Radiation Oncology department of the University Hospital Erlangen (UKER). The classification is realized via the maximum-relevance minimal-redundancy (MRMR) technique and support vector machines (SVM). The pipeline leads to a classification with a mean area under the curve (AUC) score of 0.83 in internal cross-validation and allows a division of the cohort into two subcohorts that differ significantly in their median time to progression (low-risk metastasis (LRM): 17.3 months, high-risk metastasis (HRM): 9.6 months, p < 0.01). The classification performance is especially enhanced by the analysis of medical images from different points in time (AUC 0.53 -> AUC 0.74). The results indicate that risk stratification of BM based on radiomics and machine learning during post-SRT follow-up is possible with good accuracy and should be further pursued to personalize and improve post-SRT follow-up.

Yixing Huang, Christoph Bert, Ahmed Gomaa et al.

Due to data privacy constraints, data sharing among multiple clinical centers is restricted, which impedes the development of high performance deep learning models from multicenter collaboration. Naive weight transfer methods share intermediate model weights without raw data and hence can bypass data privacy restrictions. However, performance drops are typically observed when the model is transferred from one center to the next because of the forgetting problem. Incremental transfer learning, which combines peer-to-peer federated learning and domain incremental learning, can overcome the data privacy issue and meanwhile preserve model performance by using continual learning techniques. In this work, a conventional domain/task incremental learning framework is adapted for incremental transfer learning. A comprehensive survey on the efficacy of different regularization-based continual learning methods for multicenter collaboration is performed. The influences of data heterogeneity, classifier head setting, network optimizer, model initialization, center order, and weight transfer type have been investigated thoroughly. Our framework is publicly accessible to the research community for further development.

Yixing Huang, Fuxin Fan, Ahmed Gomaa et al.

Cone-beam computed tomography (CBCT) is widely used in interventional surgeries and radiation oncology. Due to the limited size of flat-panel detectors, anatomical structures might be missing outside the limited field-of-view (FOV), which restricts the clinical applications of CBCT systems. Recently, deep learning methods have been proposed to extend the FOV for multi-slice CT systems. However, in mobile CBCT system with a smaller FOV size, projection data is severely truncated and it is challenging for a network to restore all missing structures outside the FOV. In some applications, only certain structures outside the FOV are of interest, e.g., ribs in needle path planning for liver/lung cancer diagnosis. Therefore, a task-specific data preparation method is proposed in this work, which automatically let the network focus on structures of interest instead of all the structures. Our preliminary experiment shows that Pix2pixGAN with a conventional training has the risk to reconstruct false positive and false negative rib structures from severely truncated CBCT data, whereas Pix2pixGAN with the proposed task-specific training can reconstruct all the ribs reliably. The proposed method is promising to empower CBCT with more clinical applications.

Ahmed Gomaa, Yixing Huang, Amr Hagag et al.

Background: This research aims to improve glioblastoma survival prediction by integrating MR images, clinical and molecular-pathologic data in a transformer-based deep learning model, addressing data heterogeneity and performance generalizability. Method: We propose and evaluate a transformer-based non-linear and non-proportional survival prediction model. The model employs self-supervised learning techniques to effectively encode the high-dimensional MRI input for integration with non-imaging data using cross-attention. To demonstrate model generalizability, the model is assessed with the time-dependent concordance index (Cdt) in two training setups using three independent public test sets: UPenn-GBM, UCSF-PDGM, and RHUH-GBM, each comprising 378, 366, and 36 cases, respectively. Results: The proposed transformer model achieved promising performance for imaging as well as non-imaging data, effectively integrating both modalities for enhanced performance (UPenn-GBM test-set, imaging Cdt 0.645, multimodal Cdt 0.707) while outperforming state-of-the-art late-fusion 3D-CNN-based models. Consistent performance was observed across the three independent multicenter test sets with Cdt values of 0.707 (UPenn-GBM, internal test set), 0.672 (UCSF-PDGM, first external test set) and 0.618 (RHUH-GBM, second external test set). The model achieved significant discrimination between patients with favorable and unfavorable survival for all three datasets (logrank p 1.9\times{10}^{-8}, 9.7\times{10}^{-3}, and 1.2\times{10}^{-2}). Conclusions: The proposed transformer-based survival prediction model integrates complementary information from diverse input modalities, contributing to improved glioblastoma survival prediction compared to state-of-the-art methods. Consistent performance was observed across institutions supporting model generalizability.

Florian Putz, Marlen Haderleina, Sebastian Lettmaier et al.

Thanks to the rapidly evolving integration of LLMs into decision-support tools, a significant transformation is happening across large-scale systems. Like other medical fields, the use of LLMs such as GPT-4 is gaining increasing interest in radiation oncology as well. An attempt to assess GPT-4's performance in radiation oncology was made via a dedicated 100-question examination on the highly specialized topic of radiation oncology physics, revealing GPT-4's superiority over other LLMs. GPT-4's performance on a broader field of clinical radiation oncology is further benchmarked by the ACR Radiation Oncology In-Training (TXIT) exam where GPT-4 achieved a high accuracy of 74.57%. Its performance on re-labelling structure names in accordance with the AAPM TG-263 report has also been benchmarked, achieving above 96% accuracies. Such studies shed light on the potential of LLMs in radiation oncology. As interest in the potential and constraints of LLMs in general healthcare applications continues to rise5, the capabilities and limitations of LLMs in radiation oncology decision support have not yet been fully explored.

Yixing Huang, Zahra Khodabakhshi, Ahmed Gomaa et al.

Objectives: This work aims to explore the impact of multicenter data heterogeneity on deep learning brain metastases (BM) autosegmentation performance, and assess the efficacy of an incremental transfer learning technique, namely learning without forgetting (LWF), to improve model generalizability without sharing raw data. Materials and methods: A total of six BM datasets from University Hospital Erlangen (UKER), University Hospital Zurich (USZ), Stanford, UCSF, NYU and BraTS Challenge 2023 on BM segmentation were used for this evaluation. First, the multicenter performance of a convolutional neural network (DeepMedic) for BM autosegmentation was established for exclusive single-center training and for training on pooled data, respectively. Subsequently bilateral collaboration was evaluated, where a UKER pretrained model is shared to another center for further training using transfer learning (TL) either with or without LWF. Results: For single-center training, average F1 scores of BM detection range from 0.625 (NYU) to 0.876 (UKER) on respective single-center test data. Mixed multicenter training notably improves F1 scores at Stanford and NYU, with negligible improvement at other centers. When the UKER pretrained model is applied to USZ, LWF achieves a higher average F1 score (0.839) than naive TL (0.570) and single-center training (0.688) on combined UKER and USZ test data. Naive TL improves sensitivity and contouring accuracy, but compromises precision. Conversely, LWF demonstrates commendable sensitivity, precision and contouring accuracy. When applied to Stanford, similar performance was observed. Conclusion: Data heterogeneity results in varying performance in BM autosegmentation, posing challenges to model generalizability. LWF is a promising approach to peer-to-peer privacy-preserving model training.

Ahmed Gomaa, Yixing Huang, Pluvio Stephan et al.

Accurate differentiation of pseudoprogression (PsP) from True Progression (TP) following radiotherapy (RT) in glioblastoma (GBM) patients is crucial for optimal treatment planning. However, this task remains challenging due to the overlapping imaging characteristics of PsP and TP. This study therefore proposes a multimodal deep-learning approach utilizing complementary information from routine anatomical MR images, clinical parameters, and RT treatment planning information for improved predictive accuracy. The approach utilizes a self-supervised Vision Transformer (ViT) to encode multi-sequence MR brain volumes to effectively capture both global and local context from the high dimensional input. The encoder is trained in a self-supervised upstream task on unlabeled glioma MRI datasets from the open BraTS2021, UPenn-GBM, and UCSF-PDGM datasets to generate compact, clinically relevant representations from FLAIR and T1 post-contrast sequences. These encoded MR inputs are then integrated with clinical data and RT treatment planning information through guided cross-modal attention, improving progression classification accuracy. This work was developed using two datasets from different centers: the Burdenko Glioblastoma Progression Dataset (n = 59) for training and validation, and the GlioCMV progression dataset from the University Hospital Erlangen (UKER) (n = 20) for testing. The proposed method achieved an AUC of 75.3%, outperforming the current state-of-the-art data-driven approaches. Importantly, the proposed approach relies on readily available anatomical MRI sequences, clinical data, and RT treatment planning information, enhancing its clinical feasibility. The proposed approach addresses the challenge of limited data availability for PsP and TP differentiation and could allow for improved clinical decision-making and optimized treatment plans for GBM patients.

Ugur Dinc, Jibak Sarkar, Philipp Schubert et al.

Introduction: Large language models (LLM) have shown great potential in clinical decision support. GPT-5 is a novel LLM system that has been specifically marketed towards oncology use. Methods: Performance was assessed using two complementary benchmarks: (i) the ACR Radiation Oncology In-Training Examination (TXIT, 2021), comprising 300 multiple-choice items, and (ii) a curated set of 60 authentic radiation oncologic vignettes representing diverse disease sites and treatment indications. For the vignette evaluation, GPT-5 was instructed to generate concise therapeutic plans. Four board-certified radiation oncologists rated correctness, comprehensiveness, and hallucinations. Inter-rater reliability was quantified using Fleiss' \k{appa}. Results: On the TXIT benchmark, GPT-5 achieved a mean accuracy of 92.8%, outperforming GPT-4 (78.8%) and GPT-3.5 (62.1%). Domain-specific gains were most pronounced in Dose and Diagnosis. In the vignette evaluation, GPT-5's treatment recommendations were rated highly for correctness (mean 3.24/4, 95% CI: 3.11-3.38) and comprehensiveness (3.59/4, 95% CI: 3.49-3.69). Hallucinations were rare with no case reaching majority consensus for their presence. Inter-rater agreement was low (Fleiss' \k{appa} 0.083 for correctness), reflecting inherent variability in clinical judgment. Errors clustered in complex scenarios requiring precise trial knowledge or detailed clinical adaptation. Discussion: GPT-5 clearly outperformed prior model variants on the radiation oncology multiple-choice benchmark. Although GPT-5 exhibited favorable performance in generating real-world radiation oncology treatment recommendations, correctness ratings indicate room for further improvement. While hallucinations were infrequent, the presence of substantive errors underscores that GPT-5-generated recommendations require rigorous expert oversight before clinical implementation.

Ahmed Gomaa, Annette Schwarz, Ludwig Singer et al.

Background: Differentiating radiation necrosis (RN) from tumor progression after stereotactic radiosurgery (SRS) remains a critical challenge in brain metastases. While histopathology represents the gold standard, its invasiveness limits feasibility. Conventional supervised deep learning approaches are constrained by scarce biopsy-confirmed training data. Self-supervised learning (SSL) overcomes this by leveraging the growing availability of large-scale unlabeled brain metastases imaging datasets. Methods: In a two-phase deep learning strategy inspired by the foundation model paradigm, a Vision Transformer (ViT) was pre-trained via SSL on 10,167 unlabeled multi-source T1CE MRI sub-volumes. The pre-trained ViT was then fine-tuned for RN classification using a two-channel input (T1CE MRI and segmentation masks) on the public MOLAB dataset (n=109) using 20% of datasets as same-center held-out test set. External validation was performed on a second-center test cohort (n=28). Results: The self-supervised model achieved an AUC of 0.916 on the same-center test set and 0.764 on the second center test set, surpassing the fully supervised ViT (AUC 0.624/0.496; p=0.001/0.008) and radiomics (AUC 0.807/0.691; p=0.005/0.014). Multimodal integration further improved performance (AUC 0.947/0.821; p=0.073/0.001). Attention map visualizations enabled interpretability showing the model focused on clinically relevant lesion subregions. Conclusion: Large-scale pre-training on increasingly available unlabeled brain metastases datasets substantially improves AI model performance. A two-phase multimodal deep learning strategy achieved high accuracy in differentiating radiation necrosis from tumor progression using only routine T1CE MRI and standard clinical data, providing an interpretable, clinically accessible solution that warrants further validation.

Yixing Huang, Andreas Maier, Fuxin Fan et al.

In cone-beam X-ray transmission imaging, perspective deformation causes difficulty in direct, accurate geometric assessments of anatomical structures. In this work, the perspective deformation correction problem is formulated and addressed in a framework using two complementary (180°) views. The complementary view setting provides a practical way to identify perspectively deformed structures by assessing the deviation between the two views. It also provides bounding information and reduces uncertainty for learning perspective deformation. Two representative networks Pix2pixGAN and TransU-Net for correcting perspective deformation are investigated. Experiments on numerical bead phantom data demonstrate the advantage of complementary views over orthogonal views or a single view. They show that Pix2pixGAN as a fully convolutional network achieves better performance in polar space than Cartesian space, while TransU-Net as a transformer-based hybrid network achieves comparable performance in Cartesian space to polar space. Further study demonstrates that the trained model has certain tolerance to geometric inaccuracy within calibration accuracy. The efficacy of the proposed framework on synthetic projection images from patients' chest and head data as well as real cadaver CBCT projection data and its robustness in the presence of bulky metal implants and surgical screws indicate the promising aspects of future real applications.

Yixing Huang, Christoph Bert, Philipp Sommer et al.

Brain metastases occur frequently in patients with metastatic cancer. Early and accurate detection of brain metastases is very essential for treatment planning and prognosis in radiation therapy. To improve brain metastasis detection performance with deep learning, a custom detection loss called volume-level sensitivity-specificity (VSS) is proposed, which rates individual metastasis detection sensitivity and specificity in (sub-)volume levels. As sensitivity and precision are always a trade-off in a metastasis level, either a high sensitivity or a high precision can be achieved by adjusting the weights in the VSS loss without decline in dice score coefficient for segmented metastases. To reduce metastasis-like structures being detected as false positive metastases, a temporal prior volume is proposed as an additional input of DeepMedic. The modified network is called DeepMedic+ for distinction. Our proposed VSS loss improves the sensitivity of brain metastasis detection for DeepMedic, increasing the sensitivity from 85.3% to 97.5%. Alternatively, it improves the precision from 69.1% to 98.7%. Comparing DeepMedic+ with DeepMedic with the same VSS loss, 44.4% of the false positive metastases are reduced in the high sensitivity model and the precision reaches 99.6% for the high specificity model. The mean dice coefficient for all metastases is about 0.81. With the ensemble of the high sensitivity and high specificity models, on average only 1.5 false positive metastases per patient needs further check, while the majority of true positive metastases are confirmed. The ensemble learning is able to distinguish high confidence true positive metastases from metastases candidates that require special expert review or further follow-up, being particularly well-fit to the requirements of expert support in real clinical practice.