Lubomir M. Hadjiiski PhD, Presenter: Nothing to Disclose

Alon Z. Weizer MD, Abstract Co-Author: Nothing to Disclose

Ajjai S. Alva MD, Abstract Co-Author: Nothing to Disclose

Elaine M. Caoili MD, MS, Abstract Co-Author: Nothing to Disclose

Richard H. Cohan MD, Abstract Co-Author: Consultant, General Electric Company

Heang-Ping Chan PhD, Abstract Co-Author: Nothing to Disclose

Kenny Heekon Cha BEng, Abstract Co-Author: Nothing to Disclose

Stephen Dailey, Abstract Co-Author: Nothing to Disclose

To evaluate the accuracy of our Auto-Initialized Cascaded Level Set (AI-CALS) 3D segmentation system, the WHO and the RECIST criteria in estimation of treatment response of bladder cancer using CT scans.

The AI-CALS system is designed to extract 3D lesion boundary based on level sets. The system uses as input an approximate bounding box for the lesion of interest. With IRB approval, pre- and post-chemotherapy treatment CT scans of 20 patients with bladder cancers were collected retrospectively for this preliminary study. For all cases, cystectomy was performed after treatment and the disease outcome was available as reference standard of treatment response. 35% of patients had pT0 disease (complete response) at cystectomy. A radiologist marked 20 temporal pairs of primary site cancers and also manually outlined full 3D contours on both the pre- and post-treatment scans using a GUI. For all cancers, following WHO and RECIST criteria two radiologists measured the longest diameter and its perpendicular on the pre- and post-treatment scans. Receiver operating characteristic (ROC) analysis was performed and the area under the ROC curve (AUC) was calculated to estimate the accuracy for prediction of pT0 stage (complete response) at cystectomy by the manual (3D), AI-CALS (3D), WHO (2D), and RECIST (1D) methods.

For the 20 cancers, the average pre- and post-treatment volumes from radiologist’s segmentation were 36.0 and 17.6 cm3, respectively. The AUC for prediction of pT0 disease at cystectomy was 0.68±0.13 for the AI-CALS compared to 0.72±0.11 for the manual segmentation. The difference was not significant. Prediction of pT0 disease using the RECIST criteria by two radiologists was lower than the two 3D methods with AUCs of 0.59±0.14 and 0.66±0.12, respectively. Prediction of pT0 disease using the WHO criteria by the two radiologists had AUCs of 0.50±0.14 and 0.56±0.12, respectively, which were lower than all other methods.

The 3D pre- and post-treatment volume estimates obtained by manual radiologist segmentation and AI-CALS provided more accurate depiction of the irregular 3D tumor shapes and volume changes compared to the 1D (RECIST) and 2D (WHO) estimates.

For tumors with irregular shape such as bladder cancers the 3D automated segmentation has the potential to accurately and efficiently determine tumor volume and response to treatment.

Hadjiiski, L, Weizer, A, Alva, A, Caoili, E, Cohan, R, Chan, H, Cha, K, Dailey, S, Analysis of Treatment Response of Bladder Cancers on CT Scans: Comparison of Computerized Volume Estimation with WHO and RECIST Criteria.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13025774.html