ESTRO 2022

Session Item

Gynaecological

Session Code: 6014

Session Type: Poster (digital)

Track: Clinical

Radiomic model to predict 2ysOS in Cervical Cancer patients underwent neoadjuvant chemoradiotherapy

Giulia Panza, Italy

Presentation Number: PO-1343

Abstract

Abstract Title:

Radiomic model to predict 2ysOS in Cervical Cancer patients underwent neoadjuvant chemoradiotherapy

Authors:

Giulia Panza¹, Rosa Autorino², Davide Cusumano², Luca Boldrini², Benedetta Gui², Luca Russo², Claudio Votta³, Nicola Dinapoli³, Gabriella Ferrandina⁴, Alessia Nardangeli², Maura Campitelli², Gabriella Macchia⁵, Vincenzo Valentini³, Maria Antonietta Gambacorta³

¹Istituto di Radiologia, Università Cattolica del Sacro Cuore, Roma , 2. Dipartimento di Diagnostica per Immagini, Radioterapia Oncologica ed Ematologia, Fondazione Policlinico A. Gemelli IRCCS, Roma, rome, Italy; ²Fondazione Policlinico A. Gemelli IRCCS, Roma, 2. Dipartimento di Diagnostica per Immagini, Radioterapia Oncologica ed Ematologia, , ROMA, Italy; ³Fondazione Policlinico A. Gemelli IRCCS, Roma, 2. Dipartimento di Diagnostica per Immagini, Radioterapia Oncologica ed Ematologia, ROMA, Italy; ⁴Fondazione Policlinico A. Gemelli IRCCS, Roma, Dipartimento di Ginecologia, ROMA, Italy; ⁵Università Cattolica del Sacro Cuore, Campobasso, Gemelli Molise Hospital, Campobasso, Italy

Show Affiliations

Purpose or Objective

The aim of this study is to determine if radiomics features from T2-weighted 1.5 T magnetic resonance (MR) images could predict 2 years overall survival (2yOS), in patients with Locally Advanced Cervical Cancer (LACC) after neoadjuvant chemo-radiotherapy (NACRT).

Material and Methods

We retrospectively enrolled 175 patients from two institutions (142 for the training cohort and 33 for the validation one) with LACC diagnosis (stage from IB2 to IIIC at International Federation of Gynecology and Obstetrics), that underwent NACRT followed by radical surgery from 2005 to 2018.

A total of 1557 radiomics features belonging to four families (statistical, textural, morphological and fractal features) were extracted from pre-treatment MR images.

The ability of each feature in predicting 2yOS was quantified in terms of Wilcoxon Mann Whitney test. Among the significant features, Pearson Correlation Coefficient (PCC) was calculated to quantify the correlation among the different predictors.

A logistic regression model was calculated considering the two most significant features at the univariate analysis showing the lowest PCC value.

The predictive performance of the model created was quantified out using the area under the receiver operating characteristic curve (AUC) as target metric.

Results

Based on this method, 46 different variables showed significance (p<0.05) at the univariate analysis. The radiomic model showing the highest predictive value combined the features calculated starting from the grey level co-occurrence based features, after the application of the Laplacian of Gaussian filter at two different kernel size (0.7 and 1). Such model exhibited an AUC of 0.73 (95% Confidence interval of 0.61-0.84) in the training set and 0.91 (0.72-1) in the validation set (Fig.1), suggesting its potential clinical use.

Conclusion

The proposed radiomic model has the ability to predict 2yOS in LACC patients before undergoing NACRT. To confirm the reliability of such results and translate the use of such model in clinical practice, larger studies with a consistent external validation are mandatory.