Optimal compressed sensing reconstructions of fMRI using 2D deterministic and stochastic sampling geometries

Table 3 Dyadic phase geometry: Mean SSIM results (OFF & ON) over each image pair and median over all pairs

Sample rate	Image1	Image2	Image3	Image4	Image5	Image6	Image7	Image8	Image9	Image10	Image11	Image12	Median
62.50	0.9938	0.9935	0.9890	0.9932	0.9937	0.9928	0.9880	0.9933	0.9922	0.9921	0.9949	0.9931	0.99315
48.50	0.9839	0.9806	0.9721	0.9803	0.9842	0.9781	0.9748	0.9749	0.9773	0.9784	0.9844	0.9817	0.97935
40.60	0.9661	0.9696	0.9512	0.9657	0.9633	0.9558	0.9575	0.9608	0.9596	0.9613	0.9668	0.9658	0.9623
32.80	0.9347	0.9428	0.9207	0.9358	0.9452	0.9027	0.9223	0.9371	0.9331	0.9223	0.9518	0.9422	0.93525
28.10	0.9032	0.9277	0.8863	0.9020	0.9173	0.8918	0.9163	0.9068	0.9024	0.8963	0.9268	0.9206	0.905
26.60	0.8758	0.9097	0.8612	0.8865	0.8928	0.8537	0.8440	0.8918	0.8922	0.8678	0.9068	0.8987	0.88915
25.00	0.8798	0.8957	0.8379	0.8891	0.8815	0.8705	0.8799	0.8935	0.8764	0.8666	0.8990	0.8972	0.8807
21.90	0.8452	0.8624	0.8087	0.8569	0.8394	0.8409	0.8507	0.8559	0.8483	0.8380	0.8649	0.8653	0.8495
20.30	0.7779	0.7816	0.7852	0.7822	0.7928	0.6842	0.6782	0.8073	0.8019	0.7844	0.8179	0.8110	0.7848

Average SSIM values are computed using the leave-one out method. The median parameter values of the 12 training sets are set to the median values of the 11 images in the training set and then tested on the remaining image pair. The table shows the results from the testing phase.

ISSN: 1475-925X