Potential image-based criteria of neoadjuvant chemotherapy for colon cancer: multireaders' diagnostic performance

Abstract

Purpose The objective was to assess which image-based criteria can be best accurately determined at MDCT and which results in least overtreatment. Materials and methods A total of 110 consecutive patients, who underwent curative surgery for colon cancer, were included in this retrospective study. Five radiologists independently assessed the longitudinal diameter of cancer as well as T- and N-categories. The five image-based criteria (T3cd/T4, T3/T4, T3/T4 or N+, T3cd/T4 or N2, and T3/T4 with >= 4 cm) were evaluated in terms of diagnostic accuracy, interreader agreement, and overtreatment risk using pooled receiver-operating curve and Fleiss kappa analyses. Pathologic high-risk stage II or III was used as a reference standard for assessment of overtreatment risk. Results The diagnostic accuracy of multireaders was in the acceptable range (pooled area under curve (AUC): 0.751-0.829). T3/T4 showed the highest AUC (0.829) in terms of diagnostic accuracy. T3/T4 with >= 4 cm showed the highest kappa value (kappa = 0.695) followed by T3/T4 (kappa = 0.623), indicating substantial agreement. The other three criteria revealed moderate agreement (kappa = 0.558-0.577). In terms of overtreatment ratio, T3cd/T4 and T3cd/T4 or N2 showed relatively lower ratios (T3cd/T4, 2.2%; T3cd/T4 or N2, 2.9%), whereas T3/T4 and T3/T4 or N+ revealed higher ratios (T3/T4, 8.7%; T3/T4 or N+, 9.5%). Conclusions T3/T4 was the best criterion in terms of diagnostic accuracy. However, in terms of interreader agreement and overtreatment risk, T3/T4 with >= 4 cm and T3cd/T4 were better as potential image-based criteria of neoadjuvant chemotherapy for colon cancer.