Maybe pymol can help you, it can be used to ovelap different proteins. You can first load your modeled protein, then ovrelap the 5 modeled protein structures with it one by one. studying the RMSD will give some infromation about the similarity. ALso, you can ovrelap the 5 structures mutually. this can give you a clue about the differences among the five ones. Since I don't know by what principle you will determine the accuracy of the reuslts, I have no idea to answer your second question.

how can i find which structure of ITasser is similar to the modeled protein structure

Compare every I-TASSER model with Modeller model using TM-score or LGA.

how can i find which gives the more accurate result (ITASSER or Modeler)

Without a native structure (a "true" structure of protein you're modeling) you can only do so called Model Quality Assessment (MQAP), which gives you estimates of global and sometimes local quality of the model. There are many MQA programs (MQAPs), one of the best, if not the best is QMEAN

I like to use TopMatch to compare Protein 3D structures quickly. But I also use Dali and MAMMOTH.

SuperPose : Maiti et al 2004

The ProCKSI-Server

SSAP: Taylor & Orengo, 1989

STRUCTAL: Subbiah, Laurents & Levitt, 1993 Gerstein & Levitt 1998

DALI: Holm & Sander, 1993 ; Holm & Park, 2000

DEJAVU /LSQMAN Kleywegt: 1996

CE: Shindyalov & Bourne, 1998

SSM: Krissinel & Henrick, 2003

http://www.cgl.ucsf.edu/home/meng/grpmt/structalign.html

there's a whole page dedicated to different structure alignment software/methods on wikipedia. One of my favorite is called "MAMMOTH" Check out: http://en.wikipedia.org/wiki/Structural_alignment_software

open all 6 structure in Pymol. then use the "Fit" command between every structure of ITASSER and MODELER structure:

fit (selection), (target-selection)

Not only you can see their similarity visually but also it gives you an RMSD which lower RMSD between two structures shows more similarity at the structure level.

some really good serves also exist which do Multiple Structure Alignment at the same time: Multiprot MATRAS

for the second question if you have the Wild type existing then you can also use fit to see the difference between the 6 models and the wild type. again lower RMSD can show better results. but you need to see the bondings as well where important residues are involved. fit ( mutant and name ca ), ( wildtype and name ca )