JAFROC1
JAFROC1 is a significant improvement over JAFROC.
JAFROC1 software combines two methods for analyzing free-response data: (i) original JAFROC analysis (identical to Version 2.3a) and (ii) JAFROC1 analysis. JAFROC1 has greater statistical power than original JAFROC (still provided for continuity). In other words JAFROC1 will allow you to reliably detect smaller differences between modalities. The power advantage is especially pronounced when the number of normal cases is small relative to the number of abnormal cases. Original JAFROC cannot analyze data sets with no normal cases but JAFROC1 can.
The JAFROC1 figure-of-merit A1 is defined by
.
Since it uses the highest rated NL on all images, JAFROC1 is expected to have greater power than JAFROC, which ignores these marks on abnormal images.
The idea for JAFROC1 is contained in the original JAFROC paper (Medical Physics 2004; 31:2313-2330). In that work we observed mild null hypothesis failures of JAFROC1 which caused us to conservatively recommend standard JAFROC instead of JAFROC1. Recently I found that these failures were an artifact of the simulator used in that testing. With a more realistic (search-model) simulator JAFROC1 has the correct NH behavior.
The idea for JAFROC1 is contained in the original JAFROC paper: Chakraborty & Berbaum, Medical Physics 2004; 31:2313-2330. Unlike standard JAFROC, JAFROC1 also uses the highest rated non-lesion site on abnormal images and hence the higher power. In the 2004 paper observed mild null hypothesis failures of JAFROC1 caused us to conservatively recommend standard JAFROC instead of JAFROC1. These failures were an artifact of the simulator used in that testing. Recent work with a more realistic simulator has shown that JAFROC1 is a valid method. A pre-print of a paper describing it (and other methods for analyzing FROC data which will eventually be made available on this website) is in the distribution zip file. This paper has been accepted for publication in Academic Radiology.