Abstract 1219P
Background
To evaluate the ability of radiomic features inside and outside the tumor and the effect of the external range according to the size of the tumor in predicting 2-year relapse-free survival of operable non-small cell lung cancer (NSCLC) patients from chest CT images.
Methods
For this retrospective study, chest CT images were collected for 217 NSCLC patients (mean, 73.5 years; range 59∼88 years; 212 men and 5 women) who underwent surgical resection, including 89 adenocarcinomas and 128 squamous cell carcinomas between 2003 and 2017 at the Veterans Health Service Medical Center. First, corresponding tumor interest was identified on axial CT images by a radiologist with manual annotation, and the peritumoral regions were defined externally up to 3cm at 3mm interval from the tumor boundary. Second, 69 features based on intensity, texture and shape were extracted from the intratumoral region, 58 features based on intensity and texture were extracted from the peritumoral region, and significant features were selected to distinguish between recurrence and non-recurrence groups. Finally, patients were classified as recurrence and nonrecurrence using SVM and random forest, and the probability of survival for each group of classified patient groups was estimated using the Kaplan-Meier curve.
Results
For identifying tumor size-suitable peritumoral range, the performance was evaluated according to three tumor size groups based on 3cm and 5cm. As a result, the highest AUC of the intratumoral classifier, the peritumoral classifier, and the combined classifier were 0.58, 0.75, and 0.72, respectively, and the performance of classifying the recurrence and non-recurrence within 2 years was best considering the peritumoral features. For the peritumoral classifier, the highest AUC was 0.83 in the 6mm peritumoral region of tumor less than 3 cm, 0.92 in the 27mm peritumoral region between 3cm and 5cm tumors, and 0.8 in the 9mm peritumoral region of upper 5cm tumors.
Conclusions
The performance of the 2-year relapse-free survival prediction was best considering the radiomic features of the peritumoral region, and the appropriate range of peritumoral regions was affected by tumor size.
Clinical trial identification
Editorial acknowledgement
This research was supported by the MISP (Ministry of Science, ICT & Future Planning), Korea, under the National Program for Excellence in SW)(2016-0-00022) supervised by the IITP(Institute of Information & communications Technology Planning & Evaluation(2016-0-00022) and the Bio & Medical Technology Development Program of the National Research Foundation (NRF) funded by the Ministry of Science & ICT (2015-2015M3A9A7029725).
Legal entity responsible for the study
Boryung Pharmaceutical.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.