Thesis etd-10162023-145809 |
Link copiato negli appunti
Thesis type
Tesi di specializzazione (4 anni)
Author
CHIELLINI, MARTINA
URN
etd-10162023-145809
Thesis title
Radiomics in differential diagnosis between small renal oncocytoma and clear cells carcinoma on contrast-enhanced CT
Department
RICERCA TRASLAZIONALE E DELLE NUOVE TECNOLOGIE IN MEDICINA E CHIRURGIA
Course of study
RADIODIAGNOSTICA
Supervisors
relatore Prof. Neri, Emanuele
correlatore Prof.ssa Cioni, Dania
correlatore Prof.ssa Cioni, Dania
Keywords
- clear cell
- oncocytoma
- radiomics
- renal carcinoma
Graduation session start date
07/11/2023
Availability
Withheld
Release date
07/11/2026
Summary
Purpose: The purpose of this thesis is to differentiate small (T1a; < 4cm, limited to the kidney) renal oncocytoma (RO) and clear cell renal cell carcinoma (ccRCC) on contrast-enhanced CT using the field of Radiomics.
Material and methods: Forty-seven patients who were referred to Urology Department of our University Hospital and underwent contrast-enhanced CT before surgery between January 2016 and December 2020 were retrospectively enrolled in the study. At pathology examination 36 RCCs and 11 ROs were identified. All lesions were segmented on four CT series on the same slice: unenhanced, corticomedullary phase, nephrographic phase, and excretory phase. Radiomics features were extracted by using the Pyradiomics software for all the subjects using all the phases of CT imaging. Montecarlo Cross Validation technique was used to quantify the estimator performance and to estimate the impact of data splitting. The final model built using the hyperparameter selected with Optuna was trained again on the training set and the final performance evaluation was made on the test set.
Results: Significant results were obtained by the segmentation of the lesion on corticomedullary phase of CECT imaging; Radiomics analysis of the other phases did not produce significant results. The model has a mean AUC-ROC on test set of 0.70+/-0.14, an accuracy of 0.70+/-0.14, a sensitivity of 0.38+/-0.20, a specificity of 0.85+/-0.06 and a balanced accuracy of 0.62+/-0.13. The shuffled model has an AUC-ROM of 0.50+/-0.14, an accuracy of 0.59+/-0.17, a sensitivity of 0.21+/-0.16, a specificity of 0.79+/- 0.06 and a balanced accuracy of 0.50+/-0.11.
Conclusion: This study proves that Radiomics helps to differentiate ccRCC from RO on corticomedullary phase. This is very important given that ccRCC and RO are both hypervascular lesions.
Material and methods: Forty-seven patients who were referred to Urology Department of our University Hospital and underwent contrast-enhanced CT before surgery between January 2016 and December 2020 were retrospectively enrolled in the study. At pathology examination 36 RCCs and 11 ROs were identified. All lesions were segmented on four CT series on the same slice: unenhanced, corticomedullary phase, nephrographic phase, and excretory phase. Radiomics features were extracted by using the Pyradiomics software for all the subjects using all the phases of CT imaging. Montecarlo Cross Validation technique was used to quantify the estimator performance and to estimate the impact of data splitting. The final model built using the hyperparameter selected with Optuna was trained again on the training set and the final performance evaluation was made on the test set.
Results: Significant results were obtained by the segmentation of the lesion on corticomedullary phase of CECT imaging; Radiomics analysis of the other phases did not produce significant results. The model has a mean AUC-ROC on test set of 0.70+/-0.14, an accuracy of 0.70+/-0.14, a sensitivity of 0.38+/-0.20, a specificity of 0.85+/-0.06 and a balanced accuracy of 0.62+/-0.13. The shuffled model has an AUC-ROM of 0.50+/-0.14, an accuracy of 0.59+/-0.17, a sensitivity of 0.21+/-0.16, a specificity of 0.79+/- 0.06 and a balanced accuracy of 0.50+/-0.11.
Conclusion: This study proves that Radiomics helps to differentiate ccRCC from RO on corticomedullary phase. This is very important given that ccRCC and RO are both hypervascular lesions.
File
Nome file | Dimensione |
---|---|
The thesis is not available. |