Estimation or quantification of tumour volume? CT study on irregular phantoms

Abstract

Purpose: To compare radiologists' subjective size estimation to computerised volume quantification of tumour-like phantoms in spiral CT.

Material and Methods: Eight tubular phantoms with the inside irregularly covered with silicone (8.7–31.6 ml) were imaged. The phantoms were pairwise compared to analyse the differences in silicone volumes. The observers, 2 radiologists and 2 residents, used both subjective image analysis (2 sessions) and computerised volume quantification (1 session). Accuracy and observer agreement of both methods were calculated.

Results: Subjective size estimation was correct in 51% (mean weighted kappa, κ_qw=0.73). Using four observers' mean value (κ_qw=0.81) or median value (κ_qw=0.77) slightly improved the results. Average intra-observer agreement was better than average interobserver agreement. In computerised volume quantification 70% of all classifications were correct (mean κ_qw=0.85). The results were moderate even when every second or fourth slice were measured.

Conclusion: Subjective size estimation of irregular tumours should be repeatedly performed by the same observer, or by using the mean or median estimate of several observers. Computer-based methods are even more reliable and their use is especially recommended for film readers with limited radiological experience. Only every fourth slice may be measured without a major loss of measurement accuracy.