Publication
Prediction of Unknown Primary Carcinoma in Head and Neck Cancer using Radiomics
| Summary: | The unknown primary carcinoma in head and neck cancer (HNC) is a rare disease in which cancer cells spread to lymph nodes in the upper neck, but the place where it began is unknown. The diagnostic protocol to identify the primary tumour location is challenging and invasive. In return, radiomics, a quick, low-cost, non-invasive and repeatable method, has been demonstrated in this dissertation to be a valuable tool for diagnosing the primary tumour location in these patients. The dataset analysed comprises 400 HNC patients with unknown primary carcinoma from the National Cancer Institution of Milano. The primary tumour sites already diag- nosed were Hypopharynx and Larynx (HL; n = 38), Oral Cavity (OC; n = 63), Oropharynx (OPh; n = 162) and Nasopharynx (NPh; n = 137). In total, 265 radiomic features (includ- ing shape and size, first-order, second-order, and wavelet features) were extracted from the cervical lymph nodes segmented in MRI images. The clinical information included sex, age and HPV status. Three workflows based on radiomics and machine learning methods were developed in this project. In radiomic features analysis, three correlation thresholds (0.75, 0.80, 0.85) to remove the highly correlated features and five distinctive feature selection meth- ods were assessed. The best results were achieved by the third workflow when clinical information was included in the feature set selected by Sequential Backward Selection and trained with a Linear Support Vector Machine classifier. The highest accuracies ob- tained in predicting each tumour location were 78.8% for HL, 75.4% for OC, 71.5% for OPh and 95.2% for NPh. The percentage of unclassified patients was 0.5%. The outcomes indicate that radiomics with machine learning techniques and clinical information hold the potential to predict the primary tumour site accurately. |
|---|---|
| Main Authors: | Liu, Jiaying |
| Subject: | Radiomics Machine learning Head and neck cancer Unknown primary squamous cell carcinoma MRI Feature selection |
| Year: | 2022 |
| Country: | Portugal |
| Document type: | master thesis |
| Access type: | open access |
| Associated institution: | Universidade Nova de Lisboa |
| Language: | English |
| Origin: | Repositório Institucional da UNL |
| Summary: | The unknown primary carcinoma in head and neck cancer (HNC) is a rare disease in which cancer cells spread to lymph nodes in the upper neck, but the place where it began is unknown. The diagnostic protocol to identify the primary tumour location is challenging and invasive. In return, radiomics, a quick, low-cost, non-invasive and repeatable method, has been demonstrated in this dissertation to be a valuable tool for diagnosing the primary tumour location in these patients. The dataset analysed comprises 400 HNC patients with unknown primary carcinoma from the National Cancer Institution of Milano. The primary tumour sites already diag- nosed were Hypopharynx and Larynx (HL; n = 38), Oral Cavity (OC; n = 63), Oropharynx (OPh; n = 162) and Nasopharynx (NPh; n = 137). In total, 265 radiomic features (includ- ing shape and size, first-order, second-order, and wavelet features) were extracted from the cervical lymph nodes segmented in MRI images. The clinical information included sex, age and HPV status. Three workflows based on radiomics and machine learning methods were developed in this project. In radiomic features analysis, three correlation thresholds (0.75, 0.80, 0.85) to remove the highly correlated features and five distinctive feature selection meth- ods were assessed. The best results were achieved by the third workflow when clinical information was included in the feature set selected by Sequential Backward Selection and trained with a Linear Support Vector Machine classifier. The highest accuracies ob- tained in predicting each tumour location were 78.8% for HL, 75.4% for OC, 71.5% for OPh and 95.2% for NPh. The percentage of unclassified patients was 0.5%. The outcomes indicate that radiomics with machine learning techniques and clinical information hold the potential to predict the primary tumour site accurately. |
|---|