Key points are not available for this paper at this time.
To help computer-aided diagnosis systems effectively provide support for radiologists, a personalized human-computer collaboration decision making framework is developed. Two process-driven modes and two outcome-driven modes are designed in the framework, including the comprehensive recommendation (CR) and guided recommendation (GR) modes as well as the overall assessment recommendation (OAR) and benign-malignant assessment recommendation (BMAR) modes. Using a deep two-stage reasoning model, the appropriate mode for each radiologist is selected by statistically comparing their independent diagnostic performance with that of the four collaborative modes. The framework is validated on ultrasonic images from 210 patients. Experimental results show that the OAR and BMAR modes lead to the relative increase of 6.90% and 7.47% in diagnostic performance for three experienced radiologists, and the CR and GR modes lead to the relative increase of 7.37% and 7.95% in diagnostic performance for three inexperienced radiologists. These findings highlight the applicability of the developed framework.
Fu et al. (Thu,) studied this question.