: Background: Detecting pathological lymph nodes (LNs) is crucial for establishing the proper clinical approach in patients with head and neck cancer (HNC). Positron emission tomography with [18F] fluorodeoxyglucose (FDG PET) has high diagnostic value, although it can yield false positives since FDG-avid LNs can also occur from non-cancerous diseases. Objectives: To explore if radiomics features from FDG PET can enhance the identification of pathological lymph nodes in head and neck cancer. Materials and methods: This study was carried out on n=51 cervical lymph nodes (26 negative, 25 positive) from a cohort of n=27 subjects, and the standard of reference was fine needle aspiration cytology or excisional biopsy. An initial set of 54 IBSI-compliant radiomics features, which was subsequently reduced to 31 after redundancy elimination, was considered for the analysis. Mann-Whitney U tests were performed to compare each feature between positive and negative LNs. Classification models based on two sets of features, PETBase (SUVmax, MTV and TLG) and PETRad (radiomics features), respectively, were trained using logistic regression, support vector machines and Gaussian naïve Bayes, and their performance was compared. Accuracy was estimated via leave-one-out cross-validation. Results: We identified via univariate analysis 21 features that were statistically different between positive and negative LNs. In particular, dispersion features indicated that positive LNs had higher uptake non-uniformity than the negative ones. AUC, sensitivity, specificity and accuracy obtained with logistic regression were, respectively, 0.840, 68.0%, 89.5% and 80.4% for PETBase and 0.880, 72.0%, 90.0% and 82.4% for PETRad. The other classification models showed the same trend. Conclusions: Radiomics features from FDG PET can improve the diagnostic accuracy of LN status in HNC.
Radiomics Features from Positron Emission Tomography with [18F] Fluorodeoxyglucose Can Help Predict Cervical Nodal Status in Patients with Head and Neck Cancer / Bianconi, Francesco; Salis, Roberto; Fravolini, Mario Luca; Khan, Muhammad Usama; Filippi, Luca; Marongiu, Andrea; Nuvoli, Susanna; Spanu, Angela; Palumbo, Barbara. - In: CANCERS. - ISSN 2072-6694. - 16:22(2024). [10.3390/cancers16223759]
Radiomics Features from Positron Emission Tomography with [18F] Fluorodeoxyglucose Can Help Predict Cervical Nodal Status in Patients with Head and Neck Cancer
Salis, Roberto;Marongiu, Andrea;Nuvoli, Susanna;Spanu, Angela;Palumbo, Barbara
2024-01-01
Abstract
: Background: Detecting pathological lymph nodes (LNs) is crucial for establishing the proper clinical approach in patients with head and neck cancer (HNC). Positron emission tomography with [18F] fluorodeoxyglucose (FDG PET) has high diagnostic value, although it can yield false positives since FDG-avid LNs can also occur from non-cancerous diseases. Objectives: To explore if radiomics features from FDG PET can enhance the identification of pathological lymph nodes in head and neck cancer. Materials and methods: This study was carried out on n=51 cervical lymph nodes (26 negative, 25 positive) from a cohort of n=27 subjects, and the standard of reference was fine needle aspiration cytology or excisional biopsy. An initial set of 54 IBSI-compliant radiomics features, which was subsequently reduced to 31 after redundancy elimination, was considered for the analysis. Mann-Whitney U tests were performed to compare each feature between positive and negative LNs. Classification models based on two sets of features, PETBase (SUVmax, MTV and TLG) and PETRad (radiomics features), respectively, were trained using logistic regression, support vector machines and Gaussian naïve Bayes, and their performance was compared. Accuracy was estimated via leave-one-out cross-validation. Results: We identified via univariate analysis 21 features that were statistically different between positive and negative LNs. In particular, dispersion features indicated that positive LNs had higher uptake non-uniformity than the negative ones. AUC, sensitivity, specificity and accuracy obtained with logistic regression were, respectively, 0.840, 68.0%, 89.5% and 80.4% for PETBase and 0.880, 72.0%, 90.0% and 82.4% for PETRad. The other classification models showed the same trend. Conclusions: Radiomics features from FDG PET can improve the diagnostic accuracy of LN status in HNC.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.