Imaging biomarkers of collagen architecture on baseline biopsies in association with response in patients with head and neck squamous cell carcinoma treated with immunotherapy.
Year: 2022
Type: conference-abstract
Abstract: 6047 Background: Though immunotherapy (IO) has robust clinical efficacy, only a subset of head and neck squamous cell carcinoma (HNSCC) patients have shown to benefit from it. Collagen, a major component of the Tumor Microenvironment (TME), is involved in cancer fibrosis and collagen fiber organization has been shown to be associated with patient outcome for different cancers. In this work, we evaluated whether computationally extracted collagen fiber orientation disorder (CFOD) features from tumor related stroma regions (TS) in digitized H&E-stained slides can help in distinguishing between responders and non-responders to immunotherapy in HNSCC patients. Methods: Whole slide images (WSIs) from 43 HNSCC patients treated with immunotherapy (23 non-responders and 20 responders) were obtained from University Hospitals, Cleveland; response was defined as per RECIST v1.1. A derivative-of-Gaussian model was used to identify collagen fiber orientations in TS of the digitized WSIs. Different statistics (e.g. mean, standard deviation, skewness) of the various disorder descriptors (e.g., contrast and entropy) were calculated across all the neighborhood windows in the tumor region to generate the final CFOD feature set. Minimum Redundance Maximum Relevance algorithm was employed to identify the top 3 features, which were combined with a Naive Bayes classifier to differentiate between responders and non-responders. Cross-validation was used to validate the classifier over 100 iterations via the area under the receiver operating characteristics curve (AUC). Multivariable analysis was done using a logistic regression method to see how different parameters (gender, age, smoking status) will affect IO response. Results: WSIs of responders to immunotherapy showed more disorganized collagen fiber structure as compared to non-responders WSIs. The selected top 3 features were related to the contrast feature descriptor from the orientation co-occurrence matrix. The average AUC for the model based on CFOD was 0.77±0.24. The results for multivariable analysis are shown in Table 1. Conclusions: We demonstrated an association between the architectural organization of collagen fiber from baseline biopsies of HNSCC patients with response to immunotherapy. Furthermore, we demonstrate a classifier based on computationally derived CFOD features able to derive detailed assessments of collagen orientation. None of the parameters analyzed were significantly associated with change in IO response (p<0.05). Further validation on a larger independent cohort is warranted. Multivariable analysis for IO response. Parameter p value Gender (Male/Female) p=0.14 Age (<60 years/≥60years) p=0.19 Smoking status (Yes/No) p=0.54
Source: Journal of Clinical Oncology
Institutions: Case Western Reserve University, Cleveland Clinic, Vanderbilt University Medical Center, University Hospitals Cleveland Medical Center, University Hospitals Seidman Cancer Center, The Ohio State University Comprehensive Cancer Center – Arthur G. James Cancer Hospital and Richard J. Solove Research Institute, The Ohio State University Wexner Medical Center, The Ohio State University, University Hospitals of Cleveland
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
Funder: National Institutes of Health
OpenAlex ID: W4403090009