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

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.

Brief History of Artificial Intelligence.

This article reviews the history of artificial intelligence and introduces the reader to major events that prompted interest in the field, as well as pitfalls and challenges that have slowed its development. The purpose of this article is to provide a high-level historical perspective on the development of the field over the past decades, highlighting the potential of the field for transforming health care, but also the importance of setting realistic expectations for artificial intelligence applications to avoid repeating historical cyclical trends and a third "artificial intelligence winter."

Overview of Machine Learning Part 1: Fundamentals and Classic Approaches.

The extensive body of research and advances in machine learning (ML) and the availability of a large volume of patient data make ML a powerful tool for producing models with the potential for widespread deployment in clinical settings. This article provides an overview of the classic supervised and unsupervised ML methods as well as fundamental concepts required for understanding how to develop generalizable and high-performing ML applications. It also describes the important steps for developing a ML model and how decisions made in these steps affect model performance and ability to generalize.

Dual-Energy CT Texture Analysis With Machine Learning for the Evaluation and Characterization of Cervical Lymphadenopathy.

PURPOSE: To determine whether machine learning assisted-texture analysis of multi-energy virtual monochromatic image (VMI) datasets from dual-energy CT (DECT) can be used to differentiate metastatic head and neck squamous cell carcinoma (HNSCC) lymph nodes from lymphoma, inflammatory, or normal lymph nodes. MATERIALS AND METHODS: A retrospective evaluation of 412 cervical nodes from 5 different patient groups (50 patients in total) having undergone DECT of the neck between 2013 and 2015 was performed: (1) HNSCC with pathology proven metastatic adenopathy, (2) HNSCC with pathology proven benign nodes (controls for (1)), (3) lymphoma, (4) inflammatory, and (5) normal nodes (controls for (3) and (4)). Texture analysis was performed with TexRAD® software using two independent sets of contours to assess the impact of inter-rater variation. Two machine learning algorithms (Random Forests (RF) and Gradient Boosting Machine (GBM)) were used with independent training and testing sets and determination of accuracy, sensitivity, specificity, PPV, NPV, and AUC. RESULTS: In the independent testing (prediction) sets, the accuracy for distinguishing different groups of pathologic nodes or normal nodes ranged between 80 and 95%. The models generated using texture data extracted from the independent contour sets had substantial to almost perfect agreement. The accuracy, sensitivity, specificity, PPV, and NPV for correctly classifying a lymph node as malignant (i.e. metastatic HNSCC or lymphoma) versus benign were 92%, 91%, 93%, 95%, 87%, respectively. CONCLUSION: Machine learning assisted-DECT texture analysis can help distinguish different nodal pathology and normal nodes with a high accuracy.

Radiomics and Artificial Intelligence for Biomarker and Prediction Model Development in Oncology.

Unlabelled Image.

Head and neck squamous cell carcinoma: prediction of cervical lymph node metastasis by dual-energy CT texture analysis with machine learning.

OBJECTIVES: This study was conducted in order to evaluate a novel risk stratification model using dual-energy CT (DECT) texture analysis of head and neck squamous cell carcinoma (HNSCC) with machine learning to (1) predict associated cervical lymphadenopathy and (2) compare the accuracy of spectral versus single-energy (65 keV) texture evaluation for endpoint prediction. METHODS: Eighty-seven patients with HNSCC were evaluated. Texture feature extraction was performed on virtual monochromatic images (VMIs) at 65 keV alone or different sets of multi-energy VMIs ranging from 40 to 140 keV, in addition to iodine material decomposition maps and other clinical information. Random forests (RF) models were constructed for outcome prediction with internal cross-validation in addition to the use of separate randomly selected training (70%) and testing (30%) sets. Accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were determined for predicting positive versus negative nodal status in the neck. RESULTS: Depending on the model used and subset of patients evaluated, an accuracy, sensitivity, specificity, PPV, and NPV of up to 88, 100, 67, 83, and 100%, respectively, could be achieved using multi-energy texture analysis. Texture evaluation of VMIs at 65 keV alone or in combination with only iodine maps had a much lower accuracy. CONCLUSIONS: Multi-energy DECT texture analysis of HNSCC is superior to texture analysis of 65 keV VMIs and iodine maps alone and can be used to predict cervical nodal metastases with relatively high accuracy, providing information not currently available by expert evaluation of the primary tumor alone. KEY POINTS: • Texture features of HNSCC tumor are predictive of nodal status. • Multi-energy texture analysis is superior to analysis of datasets at a single energy. • Dual-energy CT texture analysis with machine learning can enhance noninvasive diagnostic tumor evaluation.

Spectral multi-energy CT texture analysis with machine learning for tissue classification: an investigation using classification of benign parotid tumours as a testing paradigm.

OBJECTIVE: There is a rich amount of quantitative information in spectral datasets generated from dual-energy CT (DECT). In this study, we compare the performance of texture analysis performed on multi-energy datasets to that of virtual monochromatic images (VMIs) at 65 keV only, using classification of the two most common benign parotid neoplasms as a testing paradigm. METHODS: Forty-two patients with pathologically proven Warthin tumour (n = 25) or pleomorphic adenoma (n = 17) were evaluated. Texture analysis was performed on VMIs ranging from 40 to 140 keV in 5-keV increments (multi-energy analysis) or 65-keV VMIs only, which is typically considered equivalent to single-energy CT. Random forest (RF) models were constructed for outcome prediction using separate randomly selected training and testing sets or the entire patient set. RESULTS: Using multi-energy texture analysis, tumour classification in the independent testing set had accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of 92%, 86%, 100%, 100%, and 83%, compared to 75%, 57%, 100%, 100%, and 63%, respectively, for single-energy analysis. CONCLUSIONS: Multi-energy texture analysis demonstrates superior performance compared to single-energy texture analysis of VMIs at 65 keV for classification of benign parotid tumours. KEY POINTS: • We present and validate a paradigm for texture analysis of DECT scans. • Multi-energy dataset texture analysis is superior to single-energy dataset texture analysis. • DECT texture analysis has high accura\cy for diagnosis of benign parotid tumours. • DECT texture analysis with machine learning can enhance non-invasive diagnostic tumour evaluation.

Advanced Tissue Characterization and Texture Analysis Using Dual-Energy Computed Tomography: Horizons and Emerging Applications.

In the last article of this issue, advanced analysis capabilities of DECT is reviewed, including spectral Hounsfield unit attenuation curves, virtual monochromatic images, material decomposition maps, tissue effective Z determination, and other advanced post-processing DECT tools, followed by different methods of analysis of the attenuation curves generated using DECT. The article concludes with exciting future horizons and potential applications, such as the use of the rich quantitative data in dual energy CT scans for texture or radiomic analysis and the use of machine learning methods for generation of prediction models using spectral data.

Endometrial Carcinoma: MR Imaging-based Texture Model for Preoperative Risk Stratification-A Preliminary Analysis.

Purpose To evaluate the associations among mathematical modeling with the use of magnetic resonance (MR) imaging-based texture features and deep myometrial invasion (DMI), lymphovascular space invasion (LVSI), and histologic high-grade endometrial carcinoma. Materials and Methods Institutional review board approval was obtained for this retrospective study. This study included 137 women with endometrial carcinomas measuring greater than 1 cm in maximal diameter who underwent 1.5-T MR imaging before hysterectomy between January 2011 and December 2015. Texture analysis was performed with commercial research software with manual delineation of a region of interest around the tumor on MR images (T2-weighted, diffusion-weighted, and dynamic contrast material-enhanced images and apparent diffusion coefficient maps). Areas under the receiver operating characteristic curve and diagnostic performance of random forest models determined by using a subset of the most relevant texture features were estimated and compared with those of independent and blinded visual assessments by three subspecialty radiologists. Results A total of 180 texture features were extracted and ultimately limited to 11 features for DMI, 12 for LVSI, and 16 for high-grade tumor for random forest modeling. With random forest models, areas under the receiver operating characteristic curve, sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were estimated at 0.84, 79.3%, 82.3%, 81.0%, 76.7%, and 84.4% for DMI; 0.80, 80.9%, 72.5%, 76.6%, 74.3%, and 79.4% for LVSI; and 0.83, 81.0%, 76.8%, 78.1%, 60.7%, and 90.1% for high-grade tumor, respectively. Sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of visual assessment for DMI were 84.5%, 82.3%, 83.2%, 77.7%, and 87.8% (reader 3). Conclusion The mathematical models that incorporated MR imaging-based texture features were associated with the presence of DMI, LVSI, and high-grade tumor and achieved equivalent accuracy to that of subspecialty radiologists for assessment of DMI in endometrial cancers larger than 1 cm. However, these preliminary results must be interpreted with caution until they are validated with an independent data set, because the small sample size relative to the number of features extracted may have resulted in overfitting of the models. © RSNA, 2017 Online supplemental material is available for this article.