Romeo, Valeria (2018) Applications of machine learning algorithms using texture analysis-derived features extracted from computed tomography and magnetic resonance images. [Tesi di dottorato]
Preview |
Text
romeo_valeria_31.pdf Download (4MB) | Preview |
| Item Type: | Tesi di dottorato |
|---|---|
| Resource language: | English |
| Title: | Applications of machine learning algorithms using texture analysis-derived features extracted from computed tomography and magnetic resonance images |
| Creators: | Creators Email Romeo, Valeria valeria.romeo@unina.it |
| Date: | 6 December 2018 |
| Number of Pages: | 36 |
| Institution: | Università degli Studi di Napoli Federico II |
| Department: | Scienze Biomediche Avanzate |
| Dottorato: | Scienze biomorfologiche e chirurgiche |
| Ciclo di dottorato: | 31 |
| Coordinatore del Corso di dottorato: | nome email Cuocolo, Alberto cuocolo@unina.it |
| Tutor: | nome email Brunetti, Arturo UNSPECIFIED |
| Date: | 6 December 2018 |
| Number of Pages: | 36 |
| Keywords: | Texture analysis; machine learning |
| Settori scientifico-disciplinari del MIUR: | Area 06 - Scienze mediche > MED/36 - Diagnostica per immagini e radioterapia |
| Date Deposited: | 21 Dec 2018 08:41 |
| Last Modified: | 30 Jun 2020 08:26 |
| URI: | http://www.fedoa.unina.it/id/eprint/12490 |
Collection description
Radiomics relies on post-processing images derived from diagnostic examinations such as ultrasound, computed tomography (CT), magnetic resonance (MR) or positron emission tomography, by means of appropriate created algorithms with the extraction of a big amount of data. One of the main applications of radiomics is texture analysis (TA), a post processing imaging technique that analyzes the spatial variation of pixel intensity levels within an image obtaining quantitative data reflecting image heterogeneity. Machine learning (ML) is an application of artificial intelligence for recognizing patterns that can be applied to medical images, enabling the development of algorithms that can learn and make prediction. The aim of the present work is to illustrate our experience in TA and ML field using MR and CT images acquired in patients with adrenal lesions and head and neck cancer imaging, respectively. In particular, we aimed to assess the accuracy of ML algorithms in the differential diagnosis of adrenal lesions and to predict tumor grade and nodal involvement in oropharynx and oral cavity squamocellular carcinoma using MR and CT images, respectively. According to our results, the ML algorithm using MR-derived texture features correctly classified the 80% of adrenal lesions, performing better than a senior radiologist. When applied to CT-derived texture features, the ML classifier was also useful to accurately predict tumor grade, the presence of nodal involvement and to define N stage in patients with OC and OP SCC with a diagnostic accuracy of 91.6%, 85.5% and 90%, respectively Our results support the potential use of ML software employing TA-derived features for the differential diagnosis of solid lesions as well as for the prediction of histological features and the presence of nodal metastases in oncologic patients. The proven potential of ML to provide quantitative imaging biomarkers as well as the fast development of this technique will probably lead to its clinical implementation in radiological practice.
Downloads
Downloads per month over past year
Actions (login required)
 |
View Item |
