Different results: [AMICO] VS [NODDI MATLAB toolbox]

[Hongbosherlock]

Hello @daducci. Thanks for your amazing tools.
I'm new to NODDI and I know all the necessary MATLAB functions have in fact been ported to Python.
But the results seems to be different, which one should I choose ?

thanks in advance!

Data

bvals:

0 0 0 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855 2855

bvecs:

0 0 0 -0.011752 0.379972 0.778022 -0.090014 -0.326883 -0.303206 0.783524 -0.085945 -0.003616 -0.585019 0.167681 0.345383 -0.327952 -0.766873 0.701003 -0.863187 -0.042416 -0.806994 0.9945 0.948573 -0.535154 0.352825 -0.498455 -0.12567 -0.637077 0.434296 -0.382373 0.036838 0.653691 0.09219 -0.852869 -0.315006 0.694287 -0.410925 -0.566358 0.560665 -0.622796 0.766714 0.395554 0.511789 0.188151 0.394789 -0.072274 -0.883701 0.24002 -0.610786 0.644897 0.02192 -0.985615 -0.689175 -0.646243 -0.182617 -0.289115 -0.209705 -0.262609 0.867817 -0.962208 0.614269 0.433382 0.91464 0.255154 -0.313052 -0.253414 0.62883 -0.770521 0.883636 -0.968903 0.339525 0.470884 0.052798 0.077763 -0.337783 0.930419 0.153621 -0.676869 -0.473947 -0.899917 -0.801852 0.733409 0.844374 0.931863 0.059745 0.172106 -0.645968 -0.86738 -0.50178 0.29007 0.481286 0.707631 0.535702
0 0 0 0.972516 -0.346026 0.032912 0.291948 -0.361752 0.885073 -0.534107 -0.111944 0.515994 0.591484 -0.080221 0.936379 0.513676 -0.396951 -0.306367 0.503861 0.744079 -0.04095 0.056582 0.180656 -0.78996 0.870586 -0.392531 0.138359 -0.166497 -0.844019 0.362568 0.825624 0.750743 -0.821922 0.383842 0.843099 -0.354854 0.092205 -0.685448 0.005903 0.764477 0.41667 -0.576989 -0.766846 0.504368 -0.916646 -0.641236 -0.000804 0.092407 -0.785057 0.623656 0.985101 0.151067 0.289684 0.595567 -0.264081 -0.95681 -0.86328 -0.89712 0.490909 -0.21967 0.733807 -0.640964 -0.221373 0.852577 0.67744 -0.673186 0.353786 -0.381237 0.373264 -0.091378 0.63081 0.610944 -0.980123 -0.976417 0.940335 -0.173 -0.377427 0.245451 -0.607746 0.287075 -0.568922 -0.575308 0.087612 -0.36274 -0.534086 -0.831787 0.732289 -0.467907 -0.025407 0.284018 -0.164894 -0.706397 0.261213
0 0 0 -0.232541 0.857839 -0.627375 -0.952189 -0.873088 0.353146 -0.317521 -0.989991 0.856584 -0.554886 0.982572 -0.062489 -0.792833 -0.504317 0.643999 0.032141 -0.666744 0.589138 0.088138 -0.259949 -0.299289 0.342922 0.772956 0.982377 -0.752604 0.314672 0.849903 0.563017 0.09525 0.562089 -0.35395 -0.435837 -0.626134 -0.906994 0.457602 0.828022 0.166435 -0.488401 -0.714577 -0.387324 -0.842741 -0.062461 -0.763933 -0.468052 -0.96636 0.103085 0.44177 0.170575 -0.075772 0.664169 0.477147 0.947055 -0.030439 -0.459098 0.355263 -0.076831 -0.160938 -0.290174 0.633518 0.338272 -0.45608 0.665638 0.694695 0.692393 0.510839 0.282598 0.229951 0.697711 -0.636409 -0.191238 0.201403 -0.040892 -0.323098 -0.913208 -0.693976 -0.637196 0.328234 -0.182651 0.362123 0.528542 0.007171 0.843317 -0.52774 -0.215586 0.169455 0.864622 0.913889 -0.860915 0.016188 -0.802989

Load the data log

ae.load_data(dwi_filename = "eddy_corrected_data.nii", scheme_filename = "NODDI_protocol.scheme", mask_filename = "b0_brain_mask.nii", b0_thr = 0)

-> Loading data:
        * DWI signal
                - dim    = 128 x 128 x 72 x 93
                - pixdim = 2.000 x 2.000 x 2.000
        * Acquisition scheme
                - 93 samples, 1 shells
                - 3 @ b=0 , 90 @ b=2855.0 
        * Binary mask
                - dim    = 128 x 128 x 72
                - pixdim = 2.000 x 2.000 x 2.000
                - voxels = 194310
   [ 1.0 seconds ]

-> Preprocessing:
        * Normalizing to b0... [ min=-1926982442942464.00,  mean=1052269568.00, max=3696406158114816.00 ]
        * Keeping all b0 volume(s)
   [ 2.0 seconds ]

ae.set_model("NODDI")
ae.generate_kernels( )
ae.load_kernels()
ae.fit()
ae.save_results()

Compute the response functions, then Model fit. Finally I get results.
I also processing the data separately according to NODDI Matlab Toolbox tutorial .
I visualized the results and compared them using SPM12 in MATLAB:

• NDI
  
• ODI
  
• ISOVF
  

[daducci]

Dear @Hongbosherlock ,

sorry for getting to you so late, but I have missed the notification. Your data appears to be single shell (only b=2855), but the NODDI model requires at least 2 shells, so you cannot fit it to your data as there are ambiguities in the model. The different results you get are probably due to how the two different implementations (NODDI MATLAB vs AMICO) try to resolve those ambiguities.