{"id":{"repo_id":"milano","oai_identifier":"oai:air.unimi.it:2434/1064588"},"canonical_url":"https://search.dev.ndltd.org/etd/milano/oai:air.unimi.it:2434/1064588","repository":{"repo_id":"milano","name":"Università degli Studi di Milano","base_url":"https://air.unimi.it/oai/request"},"display":{"title":"CLASSIFICATION OF PATIENTS WITH PAROTID CANCER USING DYNAMIC CONTRAST AND DIFFUSION EXAMINATIONS WITH RADIOMICS AND MACHINE LEARNING TECHNIQUES","abstract":"Objectives: The aim of this study is to evaluate the role of MRI-based radiomics analysis and machine learning using both DWI with multiples B values and dynamic contrast-enhanced T1-weighted sequences to differentiate pleomorphic adenoma (A), Warthin’s tumor (W) and malignant (M) tumors. Materials and Methods: This retrospective study involving 41 patients (19 male, 22 female, age 18–88, median 59.4) with parotid gland lesions (10 W, 15 A and 18 M) who underwent a neck MRI examination at San Paolo Hospital of Milan between April 2020 and July 2023 and histopathologically-confirmed. In total, 13 radiomics features were extracted from DWI with 11 B values and dynamic contrast-enhanced T1-weighted sequence. The intensity of association between features and type of tumor class (A, W and M) has been evaluated using three different features importance algorithms: Receiver operating characteristic (ROC) analysis, ReliefF, and Recursive Feature Elimination (RFE) that allowed the identification of the association between each considered radiomics predictor and the patient class (A, M, W), thus enabling the selection of an ensemble of optimal features to train the machine learning classification algorithm. Two different classification algorithms have been adopted to model the dataset under study, namely the Support Vector Machine (SVM) and the artificial Neural Networks (NN). Results: The SVM model provides a slightly higher overall accuracy (80%) compared to the NN one (78%). However, both models still provide a non-negligible number of false negatives, especially for class A and M, which exhibit lower sensitivity scores (below 80%), thus leaving room for further improvements to make the model more robust by increasing the dataset size and balance (as measured through the Shannon Entropy) to enhance the sensitivity score and therefore decrease the false negative rate. Analogous considerations can be done for the positive predictive value for patients in class W, which is slightly below 70% in the NN model compared to a higher value (75%) in the SVM model, thereby indicating that the SVM model could be potentially a more reliable model compared to NN one. Conclusions: Radiomics and machine learning allow a good diagnostic performance in differentiating pleomorphic adenoma, Warthin’s tumor and malignant tumor.","abstract_html":"Objectives: The aim of this study is to evaluate the role of MRI-based radiomics analysis and machine learning using both DWI with multiples B values and dynamic contrast-enhanced T1-weighted sequences to differentiate pleomorphic adenoma (A), Warthin’s tumor (W) and malignant (M) tumors. Materials and Methods: This retrospective study involving 41 patients (19 male, 22 female, age 18–88, median 59.4) with parotid gland lesions (10 W, 15 A and 18 M) who underwent a neck MRI examination at San Paolo Hospital of Milan between April 2020 and July 2023 and histopathologically-confirmed. In total, 13 radiomics features were extracted from DWI with 11 B values and dynamic contrast-enhanced T1-weighted sequence. The intensity of association between features and type of tumor class (A, W and M) has been evaluated using three different features importance algorithms: Receiver operating characteristic (ROC) analysis, ReliefF, and Recursive Feature Elimination (RFE) that allowed the identification of the association between each considered radiomics predictor and the patient class (A, M, W), thus enabling the selection of an ensemble of optimal features to train the machine learning classification algorithm. Two different classification algorithms have been adopted to model the dataset under study, namely the Support Vector Machine (SVM) and the artificial Neural Networks (NN). Results: The SVM model provides a slightly higher overall accuracy (80%) compared to the NN one (78%). However, both models still provide a non-negligible number of false negatives, especially for class A and M, which exhibit lower sensitivity scores (below 80%), thus leaving room for further improvements to make the model more robust by increasing the dataset size and balance (as measured through the Shannon Entropy) to enhance the sensitivity score and therefore decrease the false negative rate. Analogous considerations can be done for the positive predictive value for patients in class W, which is slightly below 70% in the NN model compared to a higher value (75%) in the SVM model, thereby indicating that the SVM model could be potentially a more reliable model compared to NN one. Conclusions: Radiomics and machine learning allow a good diagnostic performance in differentiating pleomorphic adenoma, Warthin’s tumor and malignant tumor.","abstract_has_math":false,"creators":["TORTORA, SILVIA"],"institution":"Università degli Studi di Milano","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["tutor: G. Carrafiello ; coordinatore: M. Del Fabbro","M. Sansone","Sansone","Mario","S. Tortora","CARRAFIELLO, GIANPAOLO","DEL FABBRO, MASSIMO"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2024,"date_issued":"2024-07-01","date_published":"2024-07-01","updated_at":"2026-07-27T20:19:05Z","subjects":["Radiomic","salivary tumor","machine learning.","Settore MED/36 - Diagnostica per Immagini e Radioterapia"],"languages":["eng"],"rights":["info:eu-repo/semantics/openAccess"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["http://dx.doi.org/10.13130/tortora-silvia_phd2024-07-01","10.13130/tortora-silvia_phd2024-07-01"],"render_values":[{"text":"http://dx.doi.org/10.13130/tortora-silvia_phd2024-07-01","href":"http://dx.doi.org/10.13130/tortora-silvia_phd2024-07-01","code":true},{"text":"10.13130/tortora-silvia_phd2024-07-01","href":"https://doi.org/10.13130/tortora-silvia_phd2024-07-01","code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/2434/1064588","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["tutor: G. Carrafiello ; coordinatore: M. Del Fabbro","M. Sansone","Sansone","Mario","S. Tortora","CARRAFIELLO, GIANPAOLO","DEL FABBRO, MASSIMO"]},{"key":"dc:creator","label":"Author","values":["TORTORA, SILVIA"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2024-07-01"]},{"key":"dc:publisher","label":"Institution","values":["Università degli Studi di Milano"]},{"key":"dc:relation","label":"Dc Relation","values":["numberofpages:29","alleditors:Sansone, Mario"]},{"key":"dc:type","label":"Dc Type","values":["info:eu-repo/semantics/doctoralThesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Radiomic","salivary tumor","machine learning.","Settore MED/36 - Diagnostica per Immagini e Radioterapia"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["info:eu-repo/semantics/openAccess"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://hdl.handle.net/2434/1064588","http://dx.doi.org/10.13130/tortora-silvia_phd2024-07-01","10.13130/tortora-silvia_phd2024-07-01"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Objectives: The aim of this study is to evaluate the role of MRI-based radiomics analysis and machine learning using both DWI with multiples B values and dynamic contrast-enhanced T1-weighted sequences to differentiate pleomorphic adenoma (A), Warthin’s tumor (W) and malignant (M) tumors. Materials and Methods: This retrospective study involving 41 patients (19 male, 22 female, age 18–88, median 59.4) with parotid gland lesions (10 W, 15 A and 18 M) who underwent a neck MRI examination at San Paolo Hospital of Milan between April 2020 and July 2023 and histopathologically-confirmed. In total, 13 radiomics features were extracted from DWI with 11 B values and dynamic contrast-enhanced T1-weighted sequence. The intensity of association between features and type of tumor class (A, W and M) has been evaluated using three different features importance algorithms: Receiver operating characteristic (ROC) analysis, ReliefF, and Recursive Feature Elimination (RFE) that allowed the identification of the association between each considered radiomics predictor and the patient class (A, M, W), thus enabling the selection of an ensemble of optimal features to train the machine learning classification algorithm. Two different classification algorithms have been adopted to model the dataset under study, namely the Support Vector Machine (SVM) and the artificial Neural Networks (NN). Results: The SVM model provides a slightly higher overall accuracy (80%) compared to the NN one (78%). However, both models still provide a non-negligible number of false negatives, especially for class A and M, which exhibit lower sensitivity scores (below 80%), thus leaving room for further improvements to make the model more robust by increasing the dataset size and balance (as measured through the Shannon Entropy) to enhance the sensitivity score and therefore decrease the false negative rate. Analogous considerations can be done for the positive predictive value for patients in class W, which is slightly below 70% in the NN model compared to a higher value (75%) in the SVM model, thereby indicating that the SVM model could be potentially a more reliable model compared to NN one. Conclusions: Radiomics and machine learning allow a good diagnostic performance in differentiating pleomorphic adenoma, Warthin’s tumor and malignant tumor."]},{"key":"dc:title","label":"Title","values":["CLASSIFICATION OF PATIENTS WITH PAROTID CANCER USING DYNAMIC CONTRAST AND DIFFUSION EXAMINATIONS WITH RADIOMICS AND MACHINE LEARNING TECHNIQUES"]}]}],"canonical_facts":{"dc:contributor":["tutor: G. Carrafiello ; coordinatore: M. Del Fabbro","M. Sansone","Sansone","Mario","S. Tortora","CARRAFIELLO, GIANPAOLO","DEL FABBRO, MASSIMO"],"dc:creator":["TORTORA, SILVIA"],"dc:date":["2024-07-01"],"dc:description":["Objectives: The aim of this study is to evaluate the role of MRI-based radiomics analysis and machine learning using both DWI with multiples B values and dynamic contrast-enhanced T1-weighted sequences to differentiate pleomorphic adenoma (A), Warthin’s tumor (W) and malignant (M) tumors. Materials and Methods: This retrospective study involving 41 patients (19 male, 22 female, age 18–88, median 59.4) with parotid gland lesions (10 W, 15 A and 18 M) who underwent a neck MRI examination at San Paolo Hospital of Milan between April 2020 and July 2023 and histopathologically-confirmed. In total, 13 radiomics features were extracted from DWI with 11 B values and dynamic contrast-enhanced T1-weighted sequence. The intensity of association between features and type of tumor class (A, W and M) has been evaluated using three different features importance algorithms: Receiver operating characteristic (ROC) analysis, ReliefF, and Recursive Feature Elimination (RFE) that allowed the identification of the association between each considered radiomics predictor and the patient class (A, M, W), thus enabling the selection of an ensemble of optimal features to train the machine learning classification algorithm. Two different classification algorithms have been adopted to model the dataset under study, namely the Support Vector Machine (SVM) and the artificial Neural Networks (NN). Results: The SVM model provides a slightly higher overall accuracy (80%) compared to the NN one (78%). However, both models still provide a non-negligible number of false negatives, especially for class A and M, which exhibit lower sensitivity scores (below 80%), thus leaving room for further improvements to make the model more robust by increasing the dataset size and balance (as measured through the Shannon Entropy) to enhance the sensitivity score and therefore decrease the false negative rate. Analogous considerations can be done for the positive predictive value for patients in class W, which is slightly below 70% in the NN model compared to a higher value (75%) in the SVM model, thereby indicating that the SVM model could be potentially a more reliable model compared to NN one. Conclusions: Radiomics and machine learning allow a good diagnostic performance in differentiating pleomorphic adenoma, Warthin’s tumor and malignant tumor."],"dc:identifier":["https://hdl.handle.net/2434/1064588","http://dx.doi.org/10.13130/tortora-silvia_phd2024-07-01","10.13130/tortora-silvia_phd2024-07-01"],"dc:language":["eng"],"dc:publisher":["Università degli Studi di Milano"],"dc:relation":["numberofpages:29","alleditors:Sansone, Mario"],"dc:rights":["info:eu-repo/semantics/openAccess"],"dc:subject":["Radiomic","salivary tumor","machine learning.","Settore MED/36 - Diagnostica per Immagini e Radioterapia"],"dc:title":["CLASSIFICATION OF PATIENTS WITH PAROTID CANCER USING DYNAMIC CONTRAST AND DIFFUSION EXAMINATIONS WITH RADIOMICS AND MACHINE LEARNING TECHNIQUES"],"dc:type":["info:eu-repo/semantics/doctoralThesis"]},"updated_at":"2026-07-27T20:19:05Z"}