A measure of difficulty is essential in computer-assisted learning for radiology education for optimization of learning processes and fitting personalized training to knowledge gaps. To investigate how Artificial Intelligence can be used to predict the level of difficulty of a lung CT scan, 10 one-step deep neural networks were trained to detect lung nodules in slices of lung CT scans from the LIDC-IDRI dataset.
With the use of Item analysis, a point biserial correlation value was calculated for 151 stacks of 10 transversal lung CT slices. The relation of the point biserial correlation to the measure of difficulty was investigated by comparison with subtlety scores of nodules, given by experienced radiologists. Ordinal categorical logistic regression analysis shows a statistically significant relationship between the calculated point biserial correlation and the subtlety scores of nodules.
Item analysis, with the use of 10 deep neural networks, is an indicator of a measure of difficulty.