ETD

Digital archive of theses discussed at the University of Pisa

 

Thesis etd-06272021-091819


Thesis type
Tesi di specializzazione (4 anni)
Author
FAVATI, BENEDETTA
URN
etd-06272021-091819
Thesis title
Breast cancer MR deep radiomics in prediction of molecular receptors status
Department
RICERCA TRASLAZIONALE E DELLE NUOVE TECNOLOGIE IN MEDICINA E CHIRURGIA
Course of study
RADIODIAGNOSTICA
Supervisors
relatore Prof. Neri, Emanuele
relatore Dott.ssa Marini, Carolina
Keywords
  • Breast
  • DCE-MRI
  • cancer subtype
  • hormonal receptors
  • radiomic
  • target therapy
Graduation session start date
26/07/2021
Availability
Withheld
Release date
26/07/2091
Summary
Breast cancer is the first leading cause of mortality among women in the world. The knowledge of the presence or absence of specific molecular receptors in the tumor such as the estrogen receptor (ER), the progesterone receptor (PR) and the human epidermal growth factor receptor 2 (HER2), is fundamental in order to set up a personalized therapy. Radiomic is an interesting field of study allowing to extract and analyze, starting from biomedical images, data that can provide decision support in clinical practice.
The aim of our study was to verify if there are radiomic features able to predict, starting from magnetic resonance images, the state of the hormone receptors of the breast tumors analyzed in order to support the decision-making and therapeutic process.
The population of this study is composed of 100 patients having breast cancer subjected to magnetic resonance enhanced by dynamic contrast (DCE-MRI) at our center. All tumors were confirmed by a biopsy and an immunohistochemical analysis was performed for each of them. A radiologist using QUIBIM PRECISION ® 2.3, identified, for each lesion, a region of interest and segmented it manually. By the information contained in the segmented ROIs were extracted 29 radiomic features. Binary logistic model and Artificial Neural Networks (ANN) technology were used to analyze the features. Diagnostic accuracy (sensitivity and specificity) obtained with ANN seems to be better than binary logistic model (based on four features as Auto Correlation Value, Sum Average Value, Gray Level Standard Deviation and D3D Value).
File