PhD Scientific Days 2022

Clinical Medicine VIII.

CT Texture Analysis can Differentiate Clear Cell from Non-Clear Cell Renal Cell Carcinomas Robustly Against Imaging Protocols

Előadó neve

Dr. Bettina, Budai

Előadó munkahelye

Department of Radiology, Medical Imaging Centre, Semmelweis University, Budapest

Előadó telefonszáma

+36307825622

Előadó e-mail címe

bettinabudai.95@gmail.com

Az előadás címe

CT Texture Analysis can Differentiate Clear Cell from Non-Clear Cell Renal Cell Carcinomas Robustly Against Imaging Protocols

Szerző(k) neve és munkahelye

Bettina Katalin Budai1, Róbert Stollmayer1, Aladár Dávid Rónaszéki1, Bence Fejér1, Pál Maurovich-Horvat1, Pál Novák Kaposi1
1 Department of Radiology, Medical Imaging Centre, Semmelweis University, Budapest

Bemutatás módja

Szóbeli

Szekció

Clinical Medicine VIII.

Language of the presentation

English

Preferred session

Clinical Medicine

Összefoglaló szövege

Introduction: Prior studies showed that radiomics parameters (RP) may help the non-invasive diagnosis of kidney tumors, however, most of them are single-center studies with no external validation.
Aims: We aimed to construct a machine learning (ML) model robust against imaging protocols that can differentiate non-clear cell from clear cell renal cell carcinomas (CCRCC).
Method: We retrospectively collected preoperative unenhanced (UN), corticomedullary (CM), and excretory (EX) phase CT scans from 209 patients diagnosed with RCCs between January 2009-April 2021. 107 RPs were extracted from the tumors on each phase. For the ML analysis, the cases were randomly split into training and test sets with a 3:1 ratio. Highly correlated RPs were filtered out based on Pearson’s correlation coefficient (r>0.95). The most predictive RPs were selected by the least absolute shrinkage and selection operator (LASSO) algorithm. A support vector classifier (SVC) was constructed to predict tumor types and its performance was evaluated on the “Kidney Tumor Segmentation 2019” (KiTS19) public dataset using receiver operating characteristic curve (ROC) analysis. The performance of the SVC was also compared with an expert radiologist’s.
Results: Our training set contained 121 CCRCCs and 38 non-CCRCCs, while our internal test set consisted of 40 CCRCCs and 13 non-CCRCCs. For external validation, we identified 50 CCRCCs and 23 non-CCRCCs from the KiTS19 dataset. After correlation-based feature selection, the LASSO algorithm identified 10 CM phase predictors that were then used for training an SVC. The SVC achieved an area under the ROC curve (AUC) value, accuracy, sensitivity, and specificity of 0.83, 0.78, 0.80, and 0.74 on the external test sets, respectively. Adding UN and EX phase RPs did not further increase the performance. Meanwhile, the expert radiologist achieved similar performance with an AUC, accuracy, sensitivity, and specificity of 0.77, 0.79, 0.84, and 0.69 in the same comparison.
Conclusion: Radiomics-based ML can help the diagnostics of RCCs. The performance of our SVC is comparable with the expert radiologist’s, and it also proved to be reproducible during external validation.
Funding: Supported by the ÚNKP-21-3-I New National Excellence Program of the Ministry for Innovation and Technology from the source of the National Research, Development and Innovation Fund.

University and Doctoral School

Semmelweis University, Károly Rácz Doctoral School of Clinical Medicine

Supervisor

Pál Novák Kaposi, MD, PhD

Publication of my abstract

I give consent to the publication of my abstract on the website of the congress.

Kind

Szabad

Status

elfogadva

Accepted presentation method

szóbeli

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

5717

Start

16:30

End

16:45

Authors (legacy)

Bettina Katalin Budai1, Róbert Stollmayer1, Aladár Dávid Rónaszéki1, Bence Fejér1, Pál Maurovich-Horvat1, Pál Novák Kaposi1
1 Department of Radiology, Medical Imaging Centre, Semmelweis University, Budapest