Abstract 257P
Background
In patients with rectal cancer, a complete pathological response rate of 30% is typically anticipated following neoadjuvant treatments. Identifying patients likely to achieve complete responses can aid clinicians in considering a watch-and-wait protocol, potentially avoiding definitive surgery. Conversely, identifying patients unlikely to respond adequately to neoadjuvant treatments may spare them from toxic therapies, favoring upfront surgery. This study investigates radiomic characteristics extracted from CT scans obtained before treatment and their correlation with the assessment criteria used for evaluating treatment response.
Methods
In this retrospective cross-sectional study, 62 patients diagnosed with rectal cancer, who underwent neoadjuvant chemoradiotherapy followed by surgery, were included. Pathological response data for the patients were retrieved from their oncology records. Pretreatment CT scans were acquired from the treatment planning system, with tumor site segmentation performed prior to treatment. Subsequent preprocessing procedures, such as noise elimination and image resizing, were executed, followed by the extraction of radiomics features. Feature selection and modeling involved the application of 11 machine learning algorithms, including a DNN model, on the radiomics data.
Results
Three different approaches were utilized to investigate radiomic characteristics. Employing the LASSO method for feature selection, the SVM model exhibited the highest performance among the 11 models evaluated, achieving an AUC of 91%. The sensitivity, specificity, positive predictive value, and negative predictive value derived from this model were 0.91, 0.90, 0.91, and 0.90, respectively.
Conclusions
The impressive statistical parameters observed across all three modeling approaches in this study underscore the significance of these data and models in predicting treatment response, especially in differentiating between non-response and complete response to treatment. This ability can assist in predicting the response rate to neoadjuvant chemoradiotherapy.
Legal entity responsible for the study
The authors.
Funding
Sabzevar University of Medical Sciences, Sabzevar.
Disclosure
All authors have declared no conflicts of interest.