Objective: To predict the risk of metastatic lymph nodes and the tumor grading related to oral tongue squamous cell carcinoma (OTSCC) through the combination of clinical data with radiomics metrics by computed tomography, and to develop a supportive approach in the management of the lymphatic cervical areas, with particular attention to the early stages (T1−T2). Between March 2016 and February 2020, patients with histologically confirmed OTSCC, treated by partial glossectomy and ipsilateral laterocervical lymphadenectomy and subjected to computed tomography (CT) before surgery, were identified by two centers: 81 patients (49 female and 32 male) with 58 years as the median age (range 19–86 years). Univariate analysis with non‐parametric tests and multivariate analysis with machine learning approaches were used. Clinical, hematological parameters and radiological features extracted by CT were considered individually and in combination. All clinical parameters showed statistically significant differences (p < 0.05) for the Kruskal−Wallis test when discriminating both the tumor grading and the metastatic lymph nodes. DOI, PLR, SII, and SIRI showed an accuracy of 0.70 (ROC analysis) when identifying the tumor grading, while an accuracy ≥ 0.78 was shown by DOI, NLR, PLR, SII, and SIRI when discriminating metastatic lymph nodes. In the context of the analysis of radiomics metrics, the original_glszm_HighGrayLevelZoneEmphasis feature was selected for identifying the tumor grading (accuracy of 0.70), while the wavelet_HHH_glrlm_LowGrayLevelRunEmphasis predictor was selected for determining metastatic lymph nodes (accuracy of 0.96). Remarkable findings were also obtained when classifying patients with a machine learning approach. Radiomics features alone can predict tumor grading with an accuracy of 0.76 using a logistic regression model, while an accuracy of 0.82 can be obtained by running a CART algorithm through a combination of three clinical parameters (SIRI, DOI, and PLR) with a radiomics feature (wavelet_LLL_glszm_SizeZoneNonUniformityNormalized). In the context of predicting metastatic lymph nodes, an accuracy of 0.94 was obtained using 15 radiomics features in a logistic regression model, while both CART and CIDT achieved an asymptotic accuracy value of 1.00 using only one radiomics feature. Radiomics features and clinical parameters have an important role in identifying tumor grading and metastatic lymph nodes. Machine learning approaches can be used as an easy-to‐use tool to stratify patients with early‐stage OTSCC, based on the identification of metastatic and non‐metastatic lymph nodes.
Radiomics Metrics Combined with Clinical Data in the Surgical Management of Early‐Stage (cT1–T2 N0) Tongue Squamous Cell Carcinomas: A Preliminary Study
Abbate V.;Maglitto F.;Arena A.;Perri F.;Granata V.;Iaconetta G.;Califano L.
2022-01-01
Abstract
Objective: To predict the risk of metastatic lymph nodes and the tumor grading related to oral tongue squamous cell carcinoma (OTSCC) through the combination of clinical data with radiomics metrics by computed tomography, and to develop a supportive approach in the management of the lymphatic cervical areas, with particular attention to the early stages (T1−T2). Between March 2016 and February 2020, patients with histologically confirmed OTSCC, treated by partial glossectomy and ipsilateral laterocervical lymphadenectomy and subjected to computed tomography (CT) before surgery, were identified by two centers: 81 patients (49 female and 32 male) with 58 years as the median age (range 19–86 years). Univariate analysis with non‐parametric tests and multivariate analysis with machine learning approaches were used. Clinical, hematological parameters and radiological features extracted by CT were considered individually and in combination. All clinical parameters showed statistically significant differences (p < 0.05) for the Kruskal−Wallis test when discriminating both the tumor grading and the metastatic lymph nodes. DOI, PLR, SII, and SIRI showed an accuracy of 0.70 (ROC analysis) when identifying the tumor grading, while an accuracy ≥ 0.78 was shown by DOI, NLR, PLR, SII, and SIRI when discriminating metastatic lymph nodes. In the context of the analysis of radiomics metrics, the original_glszm_HighGrayLevelZoneEmphasis feature was selected for identifying the tumor grading (accuracy of 0.70), while the wavelet_HHH_glrlm_LowGrayLevelRunEmphasis predictor was selected for determining metastatic lymph nodes (accuracy of 0.96). Remarkable findings were also obtained when classifying patients with a machine learning approach. Radiomics features alone can predict tumor grading with an accuracy of 0.76 using a logistic regression model, while an accuracy of 0.82 can be obtained by running a CART algorithm through a combination of three clinical parameters (SIRI, DOI, and PLR) with a radiomics feature (wavelet_LLL_glszm_SizeZoneNonUniformityNormalized). In the context of predicting metastatic lymph nodes, an accuracy of 0.94 was obtained using 15 radiomics features in a logistic regression model, while both CART and CIDT achieved an asymptotic accuracy value of 1.00 using only one radiomics feature. Radiomics features and clinical parameters have an important role in identifying tumor grading and metastatic lymph nodes. Machine learning approaches can be used as an easy-to‐use tool to stratify patients with early‐stage OTSCC, based on the identification of metastatic and non‐metastatic lymph nodes.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
