Segmentation and Classification of Head and Neck Nodal Metastases and Primary Tumors in PET/CT

Abstract

The prediction of cancer characteristics, treatment planning and patient outcome from medical images generally requires tumor delineation. In Head and Neck cancer (H&N), the automatic segmentation and differentiation of primary Gross Tumor Volumes (GTVt) and malignant lymph nodes (GTVn) is a necessary step for large-scale radiomics studies to predict patient outcome such as Progression Free Survival (PFS). Detecting malignant lymph nodes is also a crucial step for Tumor-Node-Metastases (TNM) staging and to support the decision to resect the nodes. In turn, automatic TNM staging and patient outcome prediction can greatly benefit patient care by helping clinicians to find the best personalized treatment. We propose the first model to automatically individually segment GTVt and GTVn in PET/CT images. A bi-modal 3D U-Net model is trained for multi-class and multi-components segmentation on the multi-centric HECKTOR 2020 dataset containing 254 cases. The dataset has been specifically re-annotated by experts to obtain ground truth GTVn contours. The results show promising segmentation performance for the automation of radiomics pipelines and their validation on large-scale studies for which manual annotations are not available. An average test Dice Similarity Coefficients (DSC) of 0.717 is obtained for the segmentation of GTVt. The GTVn segmentation is evaluated with an aggregated DSC to account for the cases without GTVn, which is estimated at 0.729 on the test set.

Benefits of automated gross tumor volume segmentation in head and neck cancer using multi-modality information

“…This is consistent with earlier findings of Moe et al and Guo et al, reporting superior automated delineation performance for combined PET and CT input compared to single-modality input [16,17]. For the multi-modal approaches, performance of the late fusion strategy (LF) significantly outperformed the early fusion strategy (EF), which confirms previous findings [13,22]. Mean DSC for (GTVp, GTVn) respectively between automated and corrected delineations was (81%, 89%) for CT+PET-LF CNN and (69%, 77%) for CT+MRI-LF CNN.…”

Automatic gross tumor segmentation of canine head and neck cancer using deep learning and cross-species transfer learning

“…Both Groendahl et al ( 29 ) and Moe et al ( 34 ) used the same single-center HNC patients as in our present study. The lowest mean Dice scores [0.31 ( 33 ) and 0.49 ( 64 )] were reported for auto-segmentation in multi-center patient cohorts, which is generally more challenging than single-center segmentation, using wider CT window widths. Both latter studies used similarly sized image VOIs and 3D architectures, which are generally superior to their 2D counterparts, as in or present work.…”

“…Previous studies on human HNC subjects report mean validation and/or test set Dice scores in the range of 0.31–0.66 for CNN-generated auto-segmentations of the GTV based on CT images ( 29 , 33 , 34 , 64 ). The relatively large variation in reported performances is likely related to differences in image pre-processing, such as CT window settings and VOI dimensions, the composition of the datasets and/or CNN architecture.…”

“…As HNC is a heterogeneous group of cancers, many studies on human HNC subjects focus only on one anatomical primary tumor site. Specifically oropharyngeal cancer which is one of the most frequently occurring HNC sites in humans worldwide ( 2 ), or nasopharyngeal cancer which display very distinctive properties, are commonly analyzed separately ( 31 , 36 , 41 , 64 , 68 – 72 ). A similar approach could be beneficial for further analyses of auto-segmentation of the GTV in canine HNC subjects.…”

Combining Tumor Segmentation Masks with PET/CT Images and Clinical Data in a Deep Learning Framework for Improved Prognostic Prediction in Head and Neck Squamous Cell Carcinoma

“…Data from 224 HNSCC patients from multiple institutions was provided in the 2021 HECKTOR Challenge [7][8][9][10] training set. Data for these patients included co-registered 18 F-FDG PET and CT scans, clinical data (Table 1), and ground truth manual segmentations of primary tumors derived from clinical experts.…”