Deep learning-based auto-delineation of gross tumour volumes and involved nodes in PET/CT images of head and neck cancer patients

Abstract

Purpose Identification and delineation of the gross tumour and malignant nodal volume (GTV) in medical images are vital in radiotherapy. We assessed the applicability of convolutional neural networks (CNNs) for fully automatic delineation of the GTV from FDG-PET/CT images of patients with head and neck cancer (HNC). CNN models were compared to manual GTV delineations made by experienced specialists. New structure-based performance metrics were introduced to enable in-depth assessment of auto-delineation of multiple malignant structures in individual patients. Methods U-Net CNN models were trained and evaluated on images and manual GTV delineations from 197 HNC patients. The dataset was split into training, validation and test cohorts (n= 142, n = 15 and n = 40, respectively). The Dice score, surface distance metrics and the new structure-based metrics were used for model evaluation. Additionally, auto-delineations were manually assessed by an oncologist for 15 randomly selected patients in the test cohort. Results The mean Dice scores of the auto-delineations were 55%, 69% and 71% for the CT-based, PET-based and PET/CT-based CNN models, respectively. The PET signal was essential for delineating all structures. Models based on PET/CT images identified 86% of the true GTV structures, whereas models built solely on CT images identified only 55% of the true structures. The oncologist reported very high-quality auto-delineations for 14 out of the 15 randomly selected patients. Conclusions CNNs provided high-quality auto-delineations for HNC using multimodality PET/CT. The introduced structure-wise evaluation metrics provided valuable information on CNN model strengths and weaknesses for multi-structure auto-delineation.

Machine Learning Predictive Performance Evaluation of Conventional and Fuzzy Radiomics in Clinical Cancer Imaging Cohorts

“…In the context of automated, especially DL approaches, we wish to emphasize that this study does not promote a particular fuzzy delineation approach, only the concept of incorporating probability weights into standard radiomics calculations. Deep learning is a naturally probabilistic approach; however, its output delineation is routinely dichotomized by a threshold to analyze the lesions by conventional radiomics afterwards (17,18,56). This step introduces an uncertainty into the dichotomized delineation mask (58-60), and overall, results in information loss.…”

“…Recently, deep learning (DL) has been demonstrated as a powerful technique to delineate suspicious lesions in PET for subsequent analysis with e.g. radiomics and machine learning (ML) (17)(18)(19)(20). However, the common feature of all the above approaches is that they result in a binary delineation mask or volume of interest (VOI) for radiomic analysis, meaning, that a particular PET voxel is either part of the analysis or not, regardless of how certain its membership in the given VOI is.…”

“…Third, it can consider not just the given lesion, but its surroundings with appropriate weights for the analysis. Last, it can directly handle DL delineation masks that are inherently probabilistic, but are routinely dichotomized by a threshold to provide a binary mask for subsequent analysis (17,18,20).…”

Fully Automatic Quantitative Measurement of 18F-FDG PET/CT in Thymic Epithelial Tumors Using a Convolutional Neural Network

“…Patient-wise DSCs were used to assess the tumor delineation performance of the CNN model in comparison with the groundtruth VOIs. 13,14 The SUV max , MTV, and TLG were derived for both the ground truth (manual) and DL-generated (automatic) VOIs. In addition, for the tumor VOIs having at least 64 voxels, a total of 63 shape modeling and texture analysis radiomics parameters (Supplemental Table 1, Supplemental Digital Content 1, http://links.lww.com/CNM/A368) were extracted after 64 Gray-level discretization using the "PyRadiomics" package in python (version 3.7; Python Software Foundation, Wilmington, DE).…”

Influence of Semiquantitative [18F]FDG PET and Hematological Parameters on Survival in HNSCC Patients Using Neural Network Analysis

“…Machine learning (ML) has been used to predict subtypes of disease as it learns from the experience (medical images) and PET-based AI imaging helps to assess the clinical decision [15]. It is used in head and neck cancer for prediction of diagnosis, treatment response, overall survival, and defining gross tumor volume (GTV) [16,17].…”