Deep learning auto-segmentation of cervical skeletal muscle for sarcopenia analysis in patients with head and neck cancer

Naser, Mohamed A.; Wahid, Kareem A.; Grossberg, Aaron J.; Olson, Brennan; Jain, Rishab; El-Habashy, Dina; Dede, Cem; Salama, Vivian; Abobakr, Moamen; Mohamed, Abdallah S.R.; He, Renjie; Jaskari, Joel; Sahlsten, Jaakko; Kaski, Kimmo; Fuller, C.D.

doi:10.3389/fonc.2022.930432

Cited by 12 publications

(14 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Due to various exclusion criteria, such as missing pretreatment CT scans, missing clinical information, and scan artifacts, a large number of patients were excluded from our analysis, which may impact the distribution of patient characteristics. Secondly, our median dice scores were lower than those reported by Naser et al (0.90 vs. 0.95) (27). We believe this is due to the preprocessing step we implemented to account for significant differences in CT imaging parameters, such as field of views, spacings, and slice thickness, between our development cohort (MDACC) and external test cohort (BWH).…”

Section: Discussioncontrasting

confidence: 85%

“…Secondly, our median dice scores were lower than those reported by Naser et al (0.90 vs. 0.95) (27). We believe this is due to the preprocessing step we implemented to account for significant differences in CT imaging parameters, such as field of views, spacings, and slice thickness, between our development cohort (MDACC) and external test cohort (BWH).…”

Section: Discussioncontrasting

confidence: 61%

“…We followed a two-step process, which is similar to a recent study conducted by Naser et al (27), to segment the C3 SM using two separate DL models. However, our methods differ significantly.…”

Section: Discussionmentioning

confidence: 99%

See 2 more Smart Citations

Fully-automated sarcopenia assessment in head and neck cancer: development and external validation of a deep learning pipeline

Saraf

Ravipati

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

PurposeSarcopenia is an established prognostic factor in patients diagnosed with head and neck squamous cell carcinoma (HNSCC). The quantification of sarcopenia assessed by imaging is typically achieved through the skeletal muscle index (SMI), which can be derived from cervical neck skeletal muscle (SM) segmentation and cross-sectional area. However, manual SM segmentation is labor-intensive, prone to inter-observer variability, and impractical for large-scale clinical use. To overcome this challenge, we have developed and externally validated a fully-automated image-based deep learning (DL) platform for cervical vertebral SM segmentation and SMI calculation, and evaluated the relevance of this with survival and toxicity outcomes.Materials and Methods899 patients diagnosed as having HNSCC with CT scans from multiple institutes were included, with 335 cases utilized for training, 96 for validation, 48 for internal testing and 393 for external testing. Ground truth single-slice segmentations of SM at the C3 vertebra level were manually generated by experienced radiation oncologists. To develop an efficient method of segmenting the SM, a multi-stage DL pipeline was implemented, consisting of a 2D convolutional neural network (CNN) to select the middle slice of C3 section and a 2D U-Net to segment SM areas. The model performance was evaluated using the Dice Similarity Coefficient (DSC) as the primary metric for the internal test set, and for the external test set the quality of automated segmentation was assessed manually by two experienced radiation oncologists. The L3 skeletal muscle area (SMA) and SMI were then calculated from the C3 cross sectional area (CSA) of the auto-segmented SM. Finally, established SMI cut-offs were used to perform further analyses to assess the correlation with survival and toxicity endpoints in the external institution with univariable and multivariable Cox regression.ResultsDSCs for validation set (n = 96) and internal test set (n = 48) were 0.90 (95% CI: 0.90 – 0.91) and 0.90 (95% CI: 0.89 - 0.91), respectively. The predicted CSA is highly correlated with the ground-truth CSA in both validation (r = 0.99,p< 0.0001) and test sets (r = 0.96,p< 0.0001). In the external test set (n = 377), 96.2% of the SM segmentations were deemed acceptable by consensus expert review. Predicted SMA and SMI values were highly correlated with the ground-truth values, with Pearson r ≥ 0.99 (p < 0.0001) for both the female and male patients in all datasets. Sarcopenia was associated with worse OS (HR 2.05 [95% CI 1.04 - 4.04], p = 0.04) and longer PEG tube duration (median 162 days vs. 134 days, HR 1.51 [95% CI 1.12 - 2.08], p = 0.006 in multivariate analysis.ConclusionWe developed and externally validated a fully-automated platform that strongly correlates with imaging-assessed sarcopenia in patients with H&N cancer that correlates with survival and toxicity outcomes. This study constitutes a significant stride towards the integration of sarcopenia assessment into decision-making for individuals diagnosed with HNSCC.SUMMARY STATEMENTIn this study, we developed and externally validated a deep learning model to investigate the impact of sarcopenia, defined as the loss of skeletal muscle mass, on patients with head and neck squamous cell carcinoma (HNSCC) undergoing radiotherapy. We demonstrated an efficient, fully-automated deep learning pipeline that can accurately segment C3 skeletal muscle area, calculate cross-sectional area, and derive a skeletal muscle index to diagnose sarcopenia from a standard of care CT scan. In multi-institutional data, we found that pre-treatment sarcopenia was associated with significantly reduced overall survival and an increased risk of adverse events. Given the increased vulnerability of patients with HNSCC, the assessment of sarcopenia prior to radiotherapy may aid in informed treatment decision-making and serve as a predictive marker for the necessity of early supportive measures.

show abstract

Section: Discussioncontrasting

confidence: 85%

Section: Discussioncontrasting

confidence: 61%

See 1 more Smart Citation

Fully-automated sarcopenia assessment in head and neck cancer: development and external validation of a deep learning pipeline

Saraf

Ravipati

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…To evaluate the OAR segmentation performance under different defacing schemes from volumetric MRI data, a convolutional neural network architecture, 3D U-net, which has found wide success in HNC-related segmentation tasks (28)(29)(30)(31)(32)(33), was utilized. Both contractive and expansive pathways include four blocks, where each block consists of two convolutional layers with a kernel size of 3, and each convolution is followed by an instance normalization layer and a LeakyReLU activation with 0.1 negative slope.…”

Section: Deep Learning Model For Oar Segmentation Reliabilitymentioning

confidence: 99%

Segmentation stability of human head and neck cancer medical images for radiotherapy applications under de-identification conditions: benchmarking data sharing and artificial intelligence use-cases

Sahlsten

Wahid

Glerean

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Background and Purpose: Increased demand for head and neck cancer (HNC) radiotherapy data for algorithmic development has prompted an escalation of head and neck image dataset sharing. Medical images must comply with data protection requirements so that data re-use is enabled without disclosing direct or indirect identifiers of study participants. Defacing, i.e., the removal of facial features, is often considered an acceptable compromise between data protection and re-usability. While existing defacing tools have been developed by the neuroimaging community, their acceptability for preserving data that is crucial to radiotherapy applications have not been previously explored. Therefore, we systematically investigated the impact of currently available defacing algorithms on HNC organs at risk (OARs) used for radiotherapy planning. Materials and Methods: We utilized a publicly available dataset of T2-weighted magnetic resonance imaging (MRI) scans for 55 HNC patients with eight previously segmented OARs (right/left submandibular glands, right/left parotid glands, right/left level II neck lymph nodes, right/left level III neck lymph nodes). Four publicly available defacing algorithms were investigated: mri_deface, defacer, fsl_deface, and pydeface. We implemented a 5-fold cross-validation 3D U-net based OAR auto-segmentation model to perform two main experiments: 1.) comparing original and defaced data for training when evaluated on original data; 2.) using original data for training and comparing model evaluation on original and defaced data. These models were primarily investigated using the Dice similarity coefficient (DSC). The mean surface distance (MSD), and mean Hausdorff distance at 95% (MHD95) were also explored. Results: Of the studied algorithms, mri_deface and defacer were unable to produce any usable images for evaluation, while fsl_deface and pydeface were unable to remove the face for 18 and 24% of subjects, respectively. All the remaining subjects had voxels removed from all OARs. Composite OAR DSC using defaced data for model training and evaluation yielded statistically worse mean DSC performance (p ≤ 0.05) for both fsl_deface (training = 0.770, evaluation = 0.709) and pydeface (training = 0.631, evaluation = 0.453) compared to the original data (0.780). Significance testing demonstrated that the DSC and MHD95 were significantly better (p ≤ 0.05) for the model trained with the original data for both submandibular glands and for the right parotid gland compared to the model trained with fsl_deface data. Similarly, the MSD was significantly better (p ≤ 0.05) for the model trained with the original data for the left submandibular gland and both parotid glands. When trained with pydeface data, the model had significantly worse performance (p ≤ 1e-5) compared to the model trained with the original data for all structures and metrics. Conclusion: We demonstrate the choice of defacing algorithm has a significant impact on HNC OAR auto-segmentation model development and deployment. Our work incentivizes the further development of HNC radiotherapy-specific defacing methods.

show abstract

“…Accuracy and repeatability in this study were guaranteed by employing the population psoas at the L3 vertebral level with an ICC of ≥ 0.85. Interestingly, Naser et al developed a deep learning-based autosegmentation model of cervical skeletal muscle for detecting sarcopenia in head and neck cancers (44). Several excellent nutritional risk screening instruments, except PG-SGA, were also widely employed in the clinic.…”

mentioning

confidence: 99%

Development and validation of a radiomics-based nomogram for the prediction of postoperative malnutrition in stage IB1-IIA2 cervical carcinoma

Yu¹,

Chen

et al. 2023

Front. Nutr.

View full text Add to dashboard Cite

ObjectiveIn individuals with stage IB1-IIA2 cervical cancer (CC) who received postoperative radiotherapy ± chemotherapy (PORT/CRT), the interaction between sarcopenia and malnutrition remains elusive, let alone employing a nomogram model based on radiomic features of psoas extracted at the level of the third lumbar vertebra (L3). This study was set to develop a radiomics-based nomogram model to predict malnutrition as per the Patient-Generated Subjective Global Assessment (PG-SGA) for individuals with CC.MethodsIn total, 120 individuals with CC underwent computed tomography (CT) scans before PORT/CRT. The radiomic features of psoas at L3 were obtained from non-enhanced CT images. Identification of the optimal features and construction of the rad-score formula were conducted utilizing the least absolute shrinkage and selection operator (LASSO) logistic regression to predict malnutrition in the training dataset (radiomic model). Identification of the major clinical factors in the clinical model was performed by means of binary logistic regression analysis. The radiomics-based nomogram was further developed by integrating radiomic signatures and clinical risk factors (combined model). The receiver operating characteristic (ROC) curves and decision curves analysis (DCA) were employed for the evaluation and comparison of the three models in terms of their predictive performance.ResultsTwelve radiomic features in total were chosen, and the rad-score was determined with the help of the non-zero coefficient from LASSO regression. Multivariate analysis revealed that besides rad-score, age and Eastern Cooperative Oncology Group performance status could independently predict malnutrition. As per the data of this analysis, a nomogram prediction model was constructed. The area under the ROC curves (AUC) values of the radiomic and clinical models were 0.778 and 0.847 for the training and 0.776 and 0.776 for the validation sets, respectively. An increase in the AUC was observed up to 0.972 and 0.805 in the training and validation sets, respectively, in the combined model. DCA also confirmed the clinical benefit of the combined model.ConclusionThis radiomics-based nomogram model depicted potential for use as a marker for predicting malnutrition in stage IB1-IIA2 CC patients who underwent PORT/CRT and required further investigation with a large sample size.

show abstract

Deep learning auto-segmentation of cervical skeletal muscle for sarcopenia analysis in patients with head and neck cancer

Cited by 12 publications

References 44 publications

Fully-automated sarcopenia assessment in head and neck cancer: development and external validation of a deep learning pipeline

Fully-automated sarcopenia assessment in head and neck cancer: development and external validation of a deep learning pipeline

Segmentation stability of human head and neck cancer medical images for radiotherapy applications under de-identification conditions: benchmarking data sharing and artificial intelligence use-cases

Development and validation of a radiomics-based nomogram for the prediction of postoperative malnutrition in stage IB1-IIA2 cervical carcinoma

Contact Info

Product

Resources

About