Head and Neck Tumor Segmentation for MR-Guided Applications

Magnetic resonance (MR)-guided radiation therapy (RT) is enhancing head and neck cancer (HNC) treatment through superior soft tissue contrast and longitudinal imaging capabilities. However, manual tumor segmentation remains a significant challenge, spurring interest in artificial intelligence (AI)-driven automation. To accelerate innovation in this field, we present the Head and Neck Tumor Segmentation for MR-Guided Applications (HNTS-MRG) 2024 Challenge, a satellite event of the 27th International Conference on Medical Image Computing and Computer Assisted Intervention. This challenge addresses the scarcity of large, publicly available AI-ready adaptive RT datasets in HNC and explores the potential of incorporating multi-timepoint data to enhance RT auto-segmentation performance. Participants tackled two HNC segmentation tasks: automatic delineation of primary gross tumor volume (GTVp) and gross metastatic regional lymph nodes (GTVn) on pre-RT (Task 1) and mid-RT (Task 2) T2-weighted scans. The challenge provided 150 HNC cases for training and 50 for final testing hosted on grand-challenge.org using a Docker submission framework. In total, 19 independent teams from across the world qualified by submitting both their algorithms and corresponding papers, resulting in 18 submissions for Task 1 and 15 submissions for Task 2. Evaluation using the mean aggregated Dice Similarity Coefficient showed top-performing AI methods achieved scores of 0.825 in Task 1 and 0.733 in Task 2. These results surpassed clinician interobserver variability benchmarks, marking significant strides in automated tumor segmentation for MR-guided RT applications in HNC.

Radiation therapy (RT) is a vital part of treatment for head and neck cancer, where accurate segmentation of gross tumor volume (GTV) is essential for effective treatment planning. This study investigates the use of pre-RT tumor regions and local gradient maps to enhance mid-RT tumor segmentation for head and neck cancer in MRI-guided adaptive radiotherapy. By leveraging pre-RT images and their segmentations as prior knowledge, we address the challenge of tumor localization in mid-RT segmentation. A gradient map of the tumor region from the pre-RT image is computed and applied to mid-RT images to improve tumor boundary delineation. Our approach demonstrated improved segmentation accuracy for both primary GTV (GTVp) and nodal GTV (GTVn), though performance was limited by data constraints. The final \(\hbox {DSC}_{agg}\) scores from the challenge’s test set evaluation were 0.534 for GTVp, 0.867 for GTVn, and a mean score of 0.70. This method shows potential for enhancing segmentation and treatment planning in adaptive radiotherapy. Team: DCPT-Stine’s group.

The increasing utilization of MRI in radiation therapy planning for head and neck cancer (HNC) highlights the need for precise tumor segmentation to enhance treatment efficacy and reduce side effects. This work presents segmentation models developed for the HNTS-MRG 2024 challenge by the team mic-dkfz, focusing on automated segmentation of HNC tumors from MRI images at two radiotherapy (RT) stages: before (pre-RT) and 2–4 weeks into RT (mid-RT). For Task 1 (pre-RT segmentation), we built upon the nnU-Net framework, enhancing it with the larger Residual Encoder architecture. We incorporated extensive data augmentation and applied transfer learning by pretraining the model on a diverse set of public 3D medical imaging datasets. For Task 2 (mid-RT segmentation), we adopted a longitudinal approach by integrating registered pre-RT images and their segmentations as additional inputs into the nnU-Net framework. On the test set, our models achieved mean aggregated Dice Similarity Coefficient (aggDSC) scores of 81.2 for Task 1 and 72.7 for Task 2. Especially the primary tumor (GTVp) segmentation is challenging and presents potential for further optimization. These results demonstrate the effectiveness of combining advanced architectures, transfer learning, and longitudinal data integration for automated tumor segmentation in MRI-guided adaptive radiation therapy.

Accurate segmentation of tumors in MRI-guided radiation therapy (RT) is crucial for effective treatment planning, particularly for complex malignancies such as head and neck cancer (HNC). This study presents a comparative analysis between two state-of-the-art deep learning models, nnU-Net v2 and STU-Net, for automatic tumor segmentation in pre-RT MRI images. While both models are designed for medical image segmentation, STU-Net introduces critical improvements in scalability and transferability, with parameter sizes ranging from 14 million to 1.4 billion. Leveraging large-scale pre-training on datasets such as TotalSegmentator, STU-Net captures complex and variable tumor structures more effectively. We modified the default nnU-Net v2 by adding additional convolutional layers to both the encoder and decoder, improving its performance for MRI data. Based on our experimental results, STU-Net demonstrated better performance than nnU-Net v2 in the head and neck tumor segmentation challenge. These findings suggest that integrating advanced models like STU-Net into clinical workflows could remarkably enhance the precision of RT planning, potentially improving patient outcomes. Ultimately, the performance of the fine-tuned STU-Net-B model submitted for the final evaluation phase of Task 1 in this challenge achieved a DSCagg-GTVp of 0.76, a DSCagg-GTVn of 0.85, and an overall DSCagg-mean score of 0.81, securing ninth place in the Task 1 rankings. The described solution is by team SZTU-SingularMatrix for Head and Neck Tumor Segmentation for MR-Guided Applications (HNTS-MRG) 2024 challenge. Link to the trained model weights: https://​github.​com/​Duskwang/​Weight/​releases.

Head and neck tumors and metastatic lymph nodes are crucial for treatment planning and prognostic analysis. Accurate segmentation and quantitative analysis of these structures require pixel-level annotation, making automated segmentation techniques essential for the diagnosis and treatment of head and neck cancer. In this study, we investigated the effects of multiple strategies on the segmentation of pre-radiotherapy (pre-RT) and mid-radiotherapy (mid-RT) images. For the segmentation of pre-RT images, we utilized: 1) a fully supervised learning approach, and 2) the same approach enhanced with pre-trained weights and the MixUp data augmentation technique. For mid-RT images, we introduced a novel computational-friendly network architecture that features separate encoders for mid-RT images and registered pre-RT images with their labels. The mid-RT encoder branch integrates information from pre-RT images and labels progressively during the forward propagation. We selected the highest-performing model from each fold and used their predictions to create an ensemble average for inference. In the final test, our models achieved a segmentation performance of 82.38% for pre-RT and 72.53% for mid-RT on aggregated Dice Similarity Coefficient (DSC) as HiLab. Our code is available at https://​github.​com/​WltyBY/​HNTS-MRG2024_​train_​code.

MRI-guided radiotherapy (RT) planning offers key advantages over conventional CT-based methods, including superior soft tissue contrast and the potential for daily adaptive RT due to the reduction of the radiation burden. In the Head and Neck (HN) region labor-intensive and time-consuming tumor segmentation still limits full utilization of MRI-guided adaptive RT. The HN Tumor Segmentation for MR-Guided Applications 2024 challenge (HNTS-MRG) aims to improve automatic tumor segmentation on MRI images by providing a dataset with reference annotations for the tasks of pre-RT and mid-RT planning.

In this work, we present our approach for the HNTS-MRG challenge. Based on the insights of a thorough literature review we implemented a fast and resource-efficient two-stage segmentation method using the nnU-Net architecture with residual encoders as a backbone. In our two-stage approach we use the segmentation results of a first training round to guide the sampling process for a second refinement stage. For the pre-RT task, we achieved competitive results using only the first-stage nnU-Net. For the mid-RT task, we could significantly increase the segmentation performance of the basic first stage nnU-Net by utilizing the prior knowledge of the pre-RT plan as an additional input for the second stage refinement network. As team alpinists we achieved an aggregated Dice Coefficient of 80.97 for the pre-RT and 69.84 for the mid-RT task on the online test set of the challenge. Our code and trained model weights for the two-stage nnU-Net approach with residual encoders are available at https://​github.​com/​elitap/​hntsmrg24.

Accurate segmentation of gross tumor volume (GTV) is essential for effective MRI-guided adaptive radiotherapy (MRgART) in head and neck cancer. However, manual segmentation of the GTV over the course of therapy is time-consuming and prone to interobserver variability. Deep learning (DL) has the potential to overcome these challenges by automatically delineating GTVs. In this study, our team, UW LAIR, tackled the challenges of both pre-radiotherapy (pre-RT) (Task 1) and mid-radiotherapy (mid-RT) (Task 2) tumor volume segmentation. To this end, we developed a series of DL models for longitudinal GTV segmentation. The backbone of our models for both tasks was SegResNet with deep supervision. For Task 1, we trained the model using a combined dataset of pre-RT and mid-RT MRI data, which resulted in the improved aggregated Dice similarity coefficient (DSC_agg) on a hold-out internal testing set compared to models trained solely on pre-RT MRI data. In Task 2, we introduced mask-aware attention modules, enabling pre-RT GTV masks to influence intermediate features learned from mid-RT data. This attention-based approach yielded slight improvements over the baseline method, which concatenated mid-RT MRI with pre-RT GTV masks as input. In the final testing phase, the ensemble of 10 pre-RT segmentation models achieved an average DSC_agg of 0.794, with 0.745 for primary GTV (GTVp) and 0.844 for metastatic lymph nodes (GTVn) in Task 1. For Task 2, the ensemble of 10 mid-RT segmentation models attained an average DSC_agg of 0.733, with 0.607 for GTVp and 0.859 for GTVn, leading us to achieve 1st place. In summary, we presented a collection of DL models that could facilitate GTV segmentation in MRgART, offering the potential to streamline radiation oncology workflows.

The adoption of online adaptive MR-guided radiotherapy (MRgRT) for Head and Neck Cancer (HNC) treatment faces challenges due to the complexity of manual HNC tumor delineation. This study focused on the problem of HNC tumor segmentation and investigated the effects of different preprocessing techniques, robust segmentation models, and ensembling steps on segmentation accuracy to propose an optimal solution . We contributed to the MICCAI Head and Neck Tumor Segmentation for MR-Guided Applications (HNTS-MRG) challenge which contains segmentation of HNC tumors in Task1) pre-RT and Task2) mid-RT MR images. In the internal validation phase, the most accurate results were achieved by ensembling two models trained on maximally cropped and contrast-enhanced images which yielded average volumetric Dice scores of (0.680, 0.785) and (0.493, 0.810) for (GTVp, GTVn) on pre-RT and mid-RT volumes. For the final testing phase, the models were submitted under the team’s name of “Stockholm_Trio” and the overall segmentation performance achieved aggregated Dice scores of (0.795, 0.849) and (0.553, 0.865) for pre- and mid-RT tasks, respectively. The developed models are available at https://​github.​com/​Astarakee/​miccai24

Magnetic Resonance Imaging (MRI) plays a crucial role in MRI-guided adaptive radiotherapy for head and neck cancer (HNC) due to its superior soft-tissue contrast. However, accurately segmenting the gross tumor volume (GTV), which includes both the primary tumor (GTVp) and lymph nodes (GTVn), remains challenging. Recently, two deep learning segmentation innovations have shown great promise: UMamba, which effectively captures long-range dependencies, and the nnU-Net Residual Encoder (ResEnc), which enhances feature extraction through multistage residual blocks. In this study, we integrate these strengths into a novel approach, termed ‘UMambaAdj’. Our proposed method was evaluated on the HNTS-MRG 2024 challenge test set using pre-RT T2-weighted MRI images, achieving an aggregated Dice Similarity Coefficient (\(\hbox {DSC}_{agg}\)) of 0.751 for GTVp and 0.842 for GTVn, with a mean \(\hbox {DSC}_{agg}\) of 0.796. This approach demonstrates potential for more precise tumor delineation in MRI-guided adaptive radiotherapy, ultimately improving treatment outcomes for HNC patients. Team: DCPT-Stine’s group.

Radiation therapy (RT) is essential in treating head and neck cancer (HNC), with magnetic resonance imaging(MRI)-guided RT offering superior soft tissue contrast and functional imaging. However, manual tumor segmentation is time-consuming and complex, and therefore remains a challenge. In this study, we present our solution as team TUMOR to the HNTS-MRG24 MICCAI Challenge which is focused on automated segmentation of primary gross tumor volumes (GTVp) and metastatic lymph node gross tumor volume (GTVn) in pre-RT and mid-RT MRI images. We utilized the HNTS-MRG2024 dataset, which consists of 150 MRI scans from patients diagnosed with HNC, including original and registered pre-RT and mid-RT T2-weighted images with corresponding segmentation masks for GTVp and GTVn. We employed two state-of-the-art models in deep learning, nnUNet and MedNeXt. For Task 1, we pretrained models on pre-RT registered and mid-RT images, followed by fine-tuning on original pre-RT images. For Task 2, we combined registered pre-RT images, registered pre-RT segmentation masks, and mid-RT data as a multi-channel input for training. Our solution for Task 1 achieved 1st place in the final test phase with an aggregated Dice Similarity Coefficient of 0.8254, and our solution for Task 2 ranked 8th with a score of 0.7005. The proposed solution is publicly available at Github Repository.

Radiotherapy is the preferred treatment modality for head and neck cancer (HNC). During the treatment, adaptive radiation therapy (ART) technology is commonly employed to account for changes in target volume and alterations in patient anatomy. This adaptability ensures that treatment remains precise and effective despite these physiological variations. Magnetic resonance imaging (MRI) provides higher-resolution soft tissue images, making it valuable in target delineation of HNC treatment. The delineation in ART should adhere to the same principles as those used in the initial delineation. Consequently, the contouring performed on MR images during ART should reference the earlier delineations for consistency and accuracy. To address this, we proposed a coarse-to-fine cascade framework based on 3D U-Net to segment mid-radiotherapy HNC from T2-weighted MRI. The model consists of two interconnected components: a coarse segmentation network and a fine segmentation network, both sharing the same architecture. In the coarse segmentation phase, different forms of prior information were used as input, including dilated pre-radiotherapy masks. In the fine segmentation phase, a resampling operation based on a bounding box focuses on the region of interest, refining the prediction with the mid-radiotherapy image to achieve the final segmentation. In our experiment, the final results were achieved with an aggregated Dice Similarity Coefficient (DSC) of 0.562, indicating that the prior information plays a crucial role in enhancing segmentation accuracy. (Team name: TNL_skd)

Radiation therapy (RT) plays a pivotal role in treating head and neck cancer (HNC), with MRI-guided approaches offering superior soft tissue contrast and daily adaptive capabilities that significantly enhance treatment precision while minimizing side effects. To optimize MRI-guided adaptive RT for HNC, we propose a novel two-stage model for Head and Neck Tumor Segmentation. In the first stage, we leverage a Self-Supervised 3D Student-Teacher Learning Framework, specifically utilizing the DINOv2 architecture, to learn effective representations from a limited unlabeled dataset. This approach effectively addresses the challenge posed by the scarcity of annotated data, enabling the model to generalize better in tumor identification and segmentation. In the second stage, we fine-tune an xLSTM-based UNet model that is specifically designed to capture both spatial and sequential features of tumor progression. This hybrid architecture improves segmentation accuracy by integrating temporal dependencies, making it particularly well-suited for MRI-guided adaptive RT planning in HNC. The model’s performance is rigorously evaluated on a diverse set of HNC cases, demonstrating significant improvements over state-of-the-art deep learning models in accurately segmenting tumor structures. Our proposed solution achieved an impressive mean aggregated Dice Coefficient of 0.81 for pre-RT segments and 0.65 for mid-RT segments, underscoring its effectiveness in automated segmentation tasks. This work advances the field of HNC imaging by providing a robust, generalizable solution for automated Head and Neck Tumor Segmentation, ultimately enhancing the quality of care for patients undergoing RT. Our team name is DeepLearnAI (CEMRG). The code for this work is available at https://​github.​com/​RespectKnowledge​/​SSL-based-DINOv2_​Vision-LSTM_​Head-and-Neck-Tumor_​Segmentation.

The superior soft tissue differentiation provided by MRI may enable more accurate tumor segmentation compared to CT and PET, potentially enhancing adaptive radiotherapy treatment planning. The Head and Neck Tumor Segmentation for MR-Guided Applications challenge (HNTSMRG-24) comprises two tasks: segmentation of primary gross tumor volume (GTVp) and metastatic lymph nodes (GTVn) on T2-weighted MRI volumes obtained at (1) pre-radiotherapy (pre-RT) and (2) mid-radiotherapy (mid-RT). The training dataset consists of data from 150 patients, including MRI volumes of pre-RT, mid-RT, and pre-RT registered to the corresponding mid-RT volumes. Each MRI volume is accompanied by a label mask, generated by merging independent annotations from a minimum of three experts. For both tasks, we propose adopting the nnU-Net V2 framework by the use of a 15-fold cross-validation ensemble instead of the standard number of 5 folds for increased robustness and variability. For pre-RT segmentation, we augmented the initial training data (150 pre-RT volumes and masks) with the corresponding mid-RT data. For mid-RT segmentation, we opted for a three-channel input, which, in addition to the mid-RT MRI volume, comprises the registered pre-RT MRI volume and the corresponding mask. The mean of the aggregated Dice Similarity Coefficient for GTVp and GTVn is computed on a blind test set and determines the quality of the proposed methods. These metrics determine the final ranking of methods for both tasks separately. The final blind testing (50 patients) of the methods proposed by our team, \(RUG\_UMCG\), resulted in an aggregated Dice Similarity Coefficient of 0.81 (0.77 for GTVp and 0.85 for GTVn) for Task 1 and 0.70 (0.54 for GTVp and 0.86 for GTVn) for Task 2.

Accurate segmentation of the primary gross tumor volumes and metastatic lymph nodes in head and neck cancer is crucial for radiotherapy but remains challenging due to high interobserver variability, highlighting a need for an effective auto-segmentation tool. Tumor delineation is used throughout radiotherapy for treatment planning, initially for pre-radiotherapy (pre-RT) MRI scans followed-up by mid-radiotherapy (mid-RT) during the treatment. For the pre-RT task, we propose a dual-stage 3D UNet approach using cascaded neural networks for progressive accuracy refinement. The first-stage models produce an initial binary segmentation, which is then refined with an ensemble of second-stage models for a multiclass segmentation. In Head and Neck Tumor Segmentation for MR-Guided Applications (HNTS-MRG) 2024 Task 1, we utilize a dataset consisting of pre-RT and mid-RT T2-weighted MRI scans. The method is trained using 5-fold cross-validation and evaluated as an ensemble of five coarse models and ten refinement models. Our approach (team FinoxyAI) achieves a mean aggregated Dice similarity coefficient of 0.737 on the test set. Moreover, with this metric, our dual-stage approach highlights consistent improvement in segmentation performance across all folds compared to a single-stage segmentation method.

Radiation therapy is one of the most frequently applied cancer treatments worldwide, especially in the context of head and neck cancer. Today, MRI-guided radiation therapy planning is becoming increasingly popular due to good soft tissue contrast, lack of radiation dose delivered to the patient, and the capability of performing functional imaging. However, MRI-guided radiation therapy requires segmenting of the cancer both before and during radiation therapy. So far, the segmentation was often performed manually by experienced radiologists, however, recent advances in deep learning-based segmentation suggest that it may be possible to perform the segmentation automatically. Nevertheless, the task is arguably more difficult when using MRI compared to e.g. PET-CT because even manual segmentation of head and neck cancer in MRI volumes is challenging and time-consuming. The importance of the problem motivated the researchers to organize the HNTSMRG challenge with the aim of developing the most accurate segmentation methods, both before and during MRI-guided radiation therapy. In this work, we benchmark several different state-of-the-art segmentation architectures to verify whether the recent advances in deep encoder-decoder architectures are impactful for low data regimes and low-contrast tasks like segmenting head and neck cancer in magnetic resonance images. We show that for such cases the traditional residual UNet-based method outperforms (DSC = 0.775/0.701) recent advances such as UNETR (DSC = 0.617/0.657), SwinUNETR (DSC = 0.757/0.700), or SegMamba (DSC = 0.708/0.683). The proposed method (lWM team) achieved a mean aggregated Dice score on the closed test set at the level of 0.771 and 0.707 for the pre- and mid-therapy segmentation tasks, scoring 14th and 6th place, respectively. The results suggest that proper data preparation, objective function, and preprocessing are more influential for the segmentation of head and neck cancer than deep network architecture.

The segmentation of head and neck cancer (HNC) tumors is a critical step in radiotherapy treatment planning. The development of automatic segmentation algorithms has the potential to streamline the radiation oncology process. In this work, we develop an ensemble of LinkNet networks for HNC tumor segmentation as part of the HNTS-MRG 2024 Grand Challenge. A single LinkNet network, pretrained on the Imagenet dataset, was trained for 200 epochs on the HNC dataset provided by the challenge. Eight good performing weights from the internal validation set were selected to create an ensemble of 2D networks. Specifically, each selected weight was used to generate a LinkNet architecture, resulting in eight networks whose predictions were averaged to produce the final predicted segmentation. Our experiments demonstrate that the ensemble network performs better than each individual architecture, leveraging the benefits of ensemble learning without the computational cost of training each network from scratch. In the challenge’s test set, the LinkNet Ensemble (team ECU) achieved an aggregated Dice score of 64.60% and 49.53% for metastatic lymph nodes and primary gross tumor segmentation, respectively, and a mean score of 57.06%.

Auto-segmentation of gross tumor volumes (GTVs) in head and neck cancer (HNC) using MRI-guided radiotherapy (RT) images presents a significant challenge that can greatly enhance clinical workflows in radiation oncology. In this study, we developed a novel deep learning model based on the nnUNetv2 framework, augmented with an autoencoder architecture. Our model introduces the original training images as an additional input channel and incorporates an MSE loss function to improve segmentation accuracy. The model was trained on a dataset of 150 HNC patients, with a private evaluation of 50 test patients as part of the HNTS-MRG 2024 challenge. The aggregated Dice similarity coefficient (DSCagg) for metastatic lymph nodes (GTVn) reached 0.8516, while the primary tumor (GTVp) scored 0.7318, with an average DSCagg of 0.7917 across both structures. By introducing an autoencoder output channel and combining dice loss with mean squared error (MSE) loss, the enhanced nnUNet architecture effectively learned additional image features to enhance segmentation accuracy. These findings suggest that deep learning models like our modified nnUNetv2 framework can significantly improve auto-segmentation accuracy in MRI-guided RT for HNC, contributing to more precise and efficient clinical workflows.

Tumor volume segmentation on MRI is a challenging and time-consuming process that is performed manually in typical clinical settings. This work presents an approach to automated delineation of head and neck tumors on MRI scans, developed in the context of the MICCAI Head and Neck Tumor Segmentation for MR-Guided Applications (HNTS-MRG) 2024 Challenge. Rather than designing a new, task-specific convolutional neural network, the focus of this research was to propose improvements to the configuration commonly used in medical segmentation tasks, relying solely on the traditional U-Net architecture. The empirical results presented in this article suggest the superiority of patch-wise normalization used for both training and sliding window inference. They also indicate that the performance of segmentation models can be enhanced by applying a scheduled data augmentation policy during training. Finally, it is shown that a small improvement in quality can be achieved by using Gaussian weighting to combine predictions for individual patches during sliding window inference. The model with the best configuration obtained an aggregated Dice Similarity Coefficient (DSCagg) of 0.749 in Task 1 and 0.710 in Task 2 on five cross-validation folds. The ensemble of five models (one best model per validation fold) showed consistent results on a private test set of 50 patients with an DSCagg of 0.752 in Task 1 and 0.718 in Task 2 (team name: andrei.iantsen). The source code and model weights are freely available at www.​github.​com/​iantsen/​hntsmrg.

Head and neck cancer (HNC) represents a significant global health burden, often requiring complex treatment strategies, including surgery, chemotherapy, and radiation therapy. Accurate delineation of tumor volumes is critical for effective treatment, particularly in MR-guided interventions, where soft tissue contrast enhances visualization of tumor boundaries. Manual segmentation of gross tumor volumes (GTV) is labor intensive, time-consuming and prone to variability, motivating the development of automated segmentation techniques. Convolutional neural networks (CNNs) have emerged as powerful tools in this task, offering significant improvements in speed and consistency. In this study, we participated as Team Pocket in Task 1 of the HNTS-MRG 2024 Grand Challenge, which focuses on the segmentation of gross tumor volumes of the primary tumor (GTVp) and the nodal tumor (GTVn) in pre-radiotherapy MR images for HNC. We evaluated the application of PocketNet, a lightweight CNN architecture, for this task. Results for the final test phase of the challenge show that PocketNet achieved an aggregated Dice Sorensen Coefficient (DSCagg) of 0.808 for GTVn and 0.732 for GTVp, with an overall mean performance of 0.77. These findings demonstrate the potential of PocketNet as an efficient and accurate solution for automated tumor segmentation in MR-guided HNC treatment workflows, with opportunities for further optimization to enhance performance.

This article explores the potential of deep learning technologies for the automated identification and delineation of primary tumor volumes (GTVp) and metastatic lymph nodes (GTVn) in radiation therapy planning, specifically using MRI data. Utilizing the high-quality dataset provided by the 2024 MICCAI Head and Neck Tumor Segmentation Challenge, this study employs the 3DnnU-Net model for automatic tumor segmentation. Our experiments revealed that the model performs poorly with high background ratios, which prompted a retraining with selected data of specific background ratios to improve segmentation performance . The results demonstrate that the model performs well on data with low background ratios, but optimization is still needed for high background ratios. Additionally, the model shows better performance in segmenting GTVn compared to GTVp, with DSCagg scores of 0.6381 and 0.8064 for Task 1 and Task 2, respectively, during the final test phase. Future work will focus on optimizing the model and adjusting the network architecture, aiming to enhance the segmentation of GTVp while maintaining the effectiveness of GTVn segmentation to increase accuracy and reliability in clinical applications.

Automated segmentation of gross tumor volumes (GTVp) and lymph nodes (GTVn) in head and neck cancer using MRI presents a critical challenge with significant potential to enhance radiation oncology workflows. In this study, we developed a deep learning pipeline based on the SegResNet architecture, integrated into the Auto3DSeg framework, to achieve fully-automated segmentation on pre-treatment (pre-RT) and mid-treatment (mid-RT) MRI scans as part of the DLaBella29 team submission to the HNTS-MRG 2024 challenge. For Task 1, we used an ensemble of six SegResNet models with predictions fused via weighted majority voting. The models were pre-trained on both pre-RT and mid-RT image-mask pairs, then fine-tuned on pre-RT data, without any pre-processing. For Task 2, an ensemble of five SegResNet models was employed, with predictions fused using majority voting. Pre-processing for Task 2 involved setting all voxels more than 1 cm from the registered pre-RT masks to background (value 0), followed by applying a bounding box to the image. Post-processing for both tasks included removing tumor predictions smaller than 175–200 mm³ and node predictions under 50–60 mm³. Our models achieved testing DSCagg scores of 0.72 and 0.82 for GTVn and GTVp in Task 1 (pre-RT MRI) and testing DSCagg scores of 0.81 and 0.49 for GTVn and GTVp in Task 2 (mid-RT MRI). This study underscores the feasibility and promise of deep learning-based auto-segmentation for improving clinical workflows in radiation oncology, particularly in adaptive radiotherapy. Future efforts will focus on refining mid-RT segmentation performance and further investigating the clinical implications of automated tumor delineation.