Nested block self-attention multiple resolution residual network for multiorgan segmentation from CT

BACKGROUND

Fast and accurate multiorgans segmentation from computed tomography (CT) scans is essential for radiation treatment planning. Self-attention(SA)-based deep learning methodologies provide higher accuracies than standard methods but require memory and computationally intensive calculations, which restricts their use to relatively shallow networks.

PURPOSE

Our goal was to develop and test a new computationally fast and memory-efficient bidirectional SA method called nested block self-attention (NBSA), which is applicable to shallow and deep multiorgan segmentation networks.

METHODS

A new multiorgan segmentation method combining a deep multiple resolution residual network with computationally efficient SA called nested block SA (MRRN-NBSA) was developed and evaluated to segment 18 different organs from head and neck (HN) and abdomen organs. MRRN-NBSA combines features from multiple image resolutions and feature levels with SA to extract organ-specific contextual features. Computational efficiency is achieved by using memory blocks of fixed spatial extent for SA calculation combined with bidirectional attention flow. Separate models were trained for HN (n = 238) and abdomen (n = 30) and tested on set aside open-source grand challenge data sets for HN (n = 10) using a public domain database of computational anatomy and blinded testing on 20 cases from Beyond the Cranial Vault data set with overall accuracy provided by the grand challenge website for abdominal organs. Robustness to two-rater segmentations was also evaluated for HN cases using the open-source data set. Statistical comparison of MRRN-NBSA against Unet, convolutional network-based SA using criss-cross attention (CCA), dual SA, and transformer-based (UNETR) methods was done by measuring the differences in the average Dice similarity coefficient (DSC) accuracy for all HN organs using the Kruskall-Wallis test, followed by individual method comparisons using paired, two-sided Wilcoxon-signed rank tests at 95% confidence level with Bonferroni correction used for multiple comparisons.

RESULTS

MRRN-NBSA produced an average high DSC of 0.88 for HN and 0.86 for the abdomen that exceeded current methods. MRRN-NBSA was more accurate than the computationally most efficient CCA (average DSC of 0.845 for HN, 0.727 for abdomen). Kruskal-Wallis test showed significant difference between evaluated methods (p=0.00025). Pair-wise comparisons showed significant differences between MRRN-NBSA than Unet (p=0.0003), CCA (p=0.030), dual (p=0.038), and UNETR methods (p=0.012) after Bonferroni correction. MRRN-NBSA produced less variable segmentations for submandibular glands (0.82 ± 0.06) compared to two raters (0.75 ± 0.31).

CONCLUSIONS

MRRN-NBSA produced more accurate multiorgan segmentations than current methods on two different public data sets. Testing on larger institutional cohorts is required to establish feasibility for clinical use.

Prospective clinical evaluation of integrating a radiation anatomist for contouring in routine radiation treatment planning

Purpose

A radiation anatomist was trained and integrated into clinical practice at a multi-site academic center. The primary objective of this quality improvement study was to determine whether a radiation anatomist improves the quality of organ-at-risk (OAR) contours, and secondarily to determine the impact on efficiency in the treatment planning process.

Methods and Materials

From March to August 2020, all patients undergoing computed tomography–based radiation planning at 2 clinics at Memorial Sloan Kettering Cancer Center were assigned using an “every other” process to either (1) OAR contouring by a radiation anatomist (intervention) or (2) contouring by the treating physician (standard of care). Blinded dosimetrists reported OAR contour quality using a 3-point scoring system based on a common clinical trial protocol deviation scale (1, acceptable; 2, minor deviation; and 3, major deviation). Physicians reported time spent contouring for all cases. Analyses included the Fisher exact test and multivariable ordinal logistic regression.

Results

There were 249 cases with data available for the primary endpoint (66% response rate). The mean OAR quality rating was 1.1 ± 0.4 for the intervention group and 1.4 ± 0.7 for the standard of care group (P < .001), with subset analysis showing a significant difference for gastrointestinal cases (n = 49; P <.001). Time from simulation to contour approval was reduced from 3 days (interquartile range [IQR], 1-6 days) in the control group to 2 days (IQR, 1-5 days) in the intervention group (P = .007). Both physicians and dosimetrists self-reported decreased time spent contouring in the intervention group compared with the control group, with a decreases of 8 minutes (17%; P < .001) and 5 minutes (50%; P = .002), respectively. Qualitative comments most often indicated edits required to bowel contours (n = 14).

Conclusions

These findings support improvements in both OAR contour quality and workflow efficiency with implementation of a radiation anatomist in routine practice. Findings could also inform development of autosegmentation by identifying disease sites and specific OARs contributing to low clinical efficiency. Future research is needed to determine the potential effect of reduced physician time spent contouring OARs on burnout.

Prospectively-validated deep learning model for segmenting swallowing and chewing structures in CT

Objective.Delineating swallowing and chewing structures aids in radiotherapy (RT) treatment planning to limit dysphagia, trismus, and speech dysfunction. We aim to develop an accurate and efficient method to automate this process.Approach.CT scans of 242 head and neck (H&N) cancer patients acquired from 2004 to 2009 at our institution were used to develop auto-segmentation models for the masseters, medial pterygoids, larynx, and pharyngeal constrictor muscle using DeepLabV3+. A cascaded framework was used, wherein models were trained sequentially to spatially constrain each structure group based on prior segmentations. Additionally, an ensemble of models, combining contextual information from axial, coronal, and sagittal views was used to improve segmentation accuracy. Prospective evaluation was conducted by measuring the amount of manual editing required in 91 H&N CT scans acquired February-May 2021.Main results. Medians and inter-quartile ranges of Dice similarity coefficients (DSC) computed on the retrospective testing set (N = 24) were 0.87 (0.85-0.89) for the masseters, 0.80 (0.79-0.81) for the medial pterygoids, 0.81 (0.79-0.84) for the larynx, and 0.69 (0.67-0.71) for the constrictor. Auto-segmentations, when compared to two sets of manual segmentations in 10 randomly selected scans, showed better agreement (DSC) with each observer than inter-observer DSC. Prospective analysis showed most manual modifications needed for clinical use were minor, suggesting auto-contouring could increase clinical efficiency. Trained segmentation models are available for research use upon request viahttps://github.com/cerr/CERR/wiki/Auto-Segmentation-models.Significance.We developed deep learning-based auto-segmentation models for swallowing and chewing structures in CT and demonstrated its potential for use in treatment planning to limit complications post-RT. To the best of our knowledge, this is the only prospectively-validated deep learning-based model for segmenting chewing and swallowing structures in CT. Segmentation models have been made open-source to facilitate reproducibility and multi-institutional research.