Site-Specific Variation in Radiomic Features of Head and Neck Squamous Cell Carcinoma and Its Impact on Machine Learning Models

Abstract

Current radiomic studies of head and neck squamous cell carcinomas (HNSCC) are typically based on datasets combining tumors from different locations, assuming that the radiomic features are similar based on histopathologic characteristics. However, molecular pathogenesis and treatment in HNSCC substantially vary across different tumor sites. It is not known if a statistical difference exists between radiomic features from different tumor sites and how they affect machine learning model performance in endpoint prediction. To answer these questions, we extracted radiomic features from contrast-enhanced neck computed tomography scans (CTs) of 605 patients with HNSCC originating from the oral cavity, oropharynx, and hypopharynx/larynx. The difference in radiomic features of tumors from these sites was assessed using statistical analyses and Random Forest classifiers on the radiomic features with 10-fold cross-validation to predict tumor sites, nodal metastasis, and HPV status. We found statistically significant differences (p-value ≤ 0.05) between the radiomic features of HNSCC depending on tumor location. We also observed that differences in quantitative features among HNSCC from different locations impact the performance of machine learning models. This suggests that radiomic features may reveal biologic heterogeneity complementary to current gold standard histopathologic evaluation. We recommend considering tumor site in radiomic studies of HNSCC.

Sparse Bayesian predictive modelling of tumour response using radiomic features

“…Specifically, HPV status is missing in 55% of the sample, with 175 missing from the LHC, 3 from the OPC and 155 from the OSCC, that is, about 95% of the total observations with site OSCC (see Figure 2). Note that HPV is considered an important predictor for lymph node metastasis (Liu et al, 2021). Therefore, given the disproportionate missingness structure, an independent site-stratified inference and prediction can result in severe loss of information and low statistical power.…”

“…Medical imaging analysis plays a central role in the personalized treatment decisions and outcome prognosis of cancer patients. For example, in head and neck squamous cell carcinoma (HNSCC), a heterogeneous malignancy constituting more than 95% of head and neck cancers (Liu et al 2021), clinical management is based on manually measured features such as the tumour size, local disease extension and presence of distant metastasis (Deschler & Day, 2008). Unfortunately, there is a considerable time cost to collecting these semantic features, since they must be quantified by a trained radiologist.…”

“…Examples of work in this area include Ji et al (2020) who implemented a lasso Cox regression using radiomics features to predict tumour recurrence after resection of early stage hepatocellular carcinoma. Vallieres et al (2017) used a random forest classifier to predict individuals who would respond to treatment for head‐and‐neck cancer, while Liu et al (2021) used a similar approach to predict nodal metastasis in patients with HNSCC.…”

Auto‐detection of necessity for MRI‐guided online adaptive replanning using a machine learning classifier

“…Figure 2c shows an example of the SSIM map (left) and its corresponding histogram (right). Six commonly used features, that is, mean, median, moment (second order), skewness, kurtosis, full width at half maximum (FWHM), 24,25 were extracted from the SSIM maps for all daily MRI sets.…”

The [18F]F-FDG PET/CT Radiomics Classifier of Histologic Subtypes and Anatomical Disease Origins across Various Malignancies: A Proof-of-Principle Study

“…A recent study investigated the site-specific variation in radiomics features of head and neck squamous cell carcinoma [8], and several further studies demonstrated the ability of radiomics features to perform histological subtype classification among several cancer types [9][10][11]. In their limited to head and neck cancer, Liu et al [8] thus suggested that the tumour site should be considered when developing radiomics-based models.…”