Training readers to improve their accuracy in grading Crohn's disease activity on MRI

To prospectively evaluate if training with direct feedback improves grading accuracy of inexperienced readers for Crohn's disease activity on magnetic resonance imaging (MRI). Thirty-one inexperienced readers assessed 25 cases as a baseline set. Subsequently, all readers received training and assessed 100 cases with direct feedback per case, randomly assigned to four sets of 25 cases. The cases in set 4 were identical to the baseline set. Grading accuracy, understaging, overstaging, mean reading times and confidence scores (scale 0-10) were compared between baseline and set 4, and between the four consecutive sets with feedback. Proportions of grading accuracy, understaging and overstaging per set were compared using logistic regression analyses. Mean reading times and confidence scores were compared by t-tests. Grading accuracy increased from 66 % (95 % CI, 56-74 %) at baseline to 75 % (95 % CI, 66-81 %) in set 4 (P = 0.003). Understaging decreased from 15 % (95 % CI, 9-23 %) to 7 % (95 % CI, 3-14 %) (P < 0.001). Overstaging did not change significantly (20 % vs 19 %). Mean reading time decreased from 6 min 37 s to 4 min 35 s (P < 0.001). Mean confidence increased from 6.90 to 7.65 (P < 0.001). During training, overall grading accuracy, understaging, mean reading times and confidence scores improved gradually. Inexperienced readers need training with at least 100 cases to achieve the literature reported grading accuracy of 75 %. (orig.)