Research Article

Deep Hash with Optimal Transport-Based Domain Adaptation for Multisite MRI Retrieval

Table 2

The mAP results achieved by different methods with different bits of hash code in retrieving data from the four sites of the ABIDE database. Data from each site are used in turn as the target domain (test set and query set) and data from the remaining sites are used as the source domain (training set and retrieval database).

2methodLeuven (target domain)NYU (target domain)UCLA (target domain)UM (target domain)
24 bits32 bits48 bits64 bits24 bits32 bits48 bits64 bits24 bits32 bits48 bits64 bits24 bits32 bits48 bits64 bits

DHN0.63620.63600.63490.65240.64230.67730.62400.70580.71290.74090.71290.71390.65280.67580.62740.6706
HashNet0.64850.60610.63690.63460.64530.63100.61180.65670.70390.74410.72640.74000.64550.62050.60120.6316
LCDSH0.63110.59840.66090.66300.59640.55620.63520.59990.71510.72560.71270.71120.58490.57620.64810.6375
DCH0.63370.64190.65020.65480.63950.66140.65400.59960.65290.67270.74230.68200.59760.62220.63650.6596
QSMIH0.65760.63500.66780.68940.64130.66440.62340.61290.66270.70170.74120.73960.58970.62540.64730.5841
DTSH0.63490.63600.63490.66700.64740.66490.59890.64720.77740.68830.71890.73790.68330.60030.58280.5891
TAH0.64860.69180.61180.64780.67380.66140.62600.65110.66690.66250.67530.70340.67790.68770.68610.7021
SH0.53430.52910.53520.52310.52530.53780.54040.53460.52400.52160.53150.52660.53770.53290.54040.5303
ITQ0.54210.55040.54580.55530.54760.53660.54650.55120.53690.54170.53830.54210.53960.55420.54390.5297
LFH0.48670.48670.48670.48670.57870.57870.57870.57870.60850.60850.60850.60850.60250.60250.60250.6025
SDH0.57890.57070.58000.60390.58540.57650.59670.58300.55390.52540.51540.52140.65200.62490.54020.5941
GTH0.52550.53120.53240.53390.54150.54060.53670.53910.53210.53000.52880.53530.53040.53210.52850.5374
MSFR (ours)0.70270.71880.74560.70540.69300.68370.69250.68800.76840.72690.75420.72360.69240.72200.75450.7209