Computer-aided texture analysis combined with experts' knowledge: Improving endoscopic celiac disease diagnosis

Table 1 Outline of the three image databases (DB-1, DB-2 and DB-3) used for experimentation

	DB-1	DB-2	DB-3
Number patches Marsh-0	280	280	280
Number patches Marsh-3	280	280	280
Number images Marsh-0	246	210	220
Number images Marsh-3	171	154	154
Number patients Marsh-0	125	82	80
Number patients Marsh-3	38	35	36
Endoscope	GIF-Q165, N180	GIF-H180	GIF-H180
Imaging technique	Traditional (white-light) imaging	Traditional (white-light) imaging	Narrow-band imaging[11]

Table 2 All individual mean classification accuracies and standard deviations (±) for all configurations. Feature extraction was performed based on local binary patterns, multi-fractal spectrum and improved Fisher vectors

Feature extraction	Data set	Expert	Patch-based				Image-based				Patient-based
			Expert diagnosis		Hybrid diagnosis		Expert diagnosis		Hybrid diagnosis		Expert diagnosis		Hybrid diagnosis
			mean	±	mean	±	mean	±	mean	±	mean	±	mean	±
LBP	DB-1	A	0.870	0.057	0.910	0.042	0.970	0.027	0.964	0.033	0.970	0.058	0.972	0.033
LBP	DB-2	A	0.857	0.042	0.910	0.040	0.957	0.026	0.955	0.038	0.988	0.047	0.999	0.005
LBP	DB-3	A	0.773	0.051	0.902	0.059	0.963	0.020	0.965	0.036	0.995	0.055	0.994	0.016
LBP	DB-1	B	0.834	0.063	0.909	0.051	0.879	0.053	0.908	0.046	0.873	0.066	0.947	0.062
LBP	DB-2	B	0.827	0.051	0.903	0.048	0.896	0.040	0.905	0.048	0.946	0.046	0.998	0.009
LBP	DB-3	B	0.627	0.058	0.883	0.057	0.818	0.040	0.913	0.047	0.832	0.070	0.960	0.064
LBP	DB-1	C	0.882	0.052	0.916	0.045	0.778	0.070	0.902	0.057	0.799	0.064	0.959	0.053
LBP	DB-2	C	0.912	0.037	0.926	0.044	0.893	0.033	0.914	0.040	0.943	0.036	0.991	0.017
LBP	DB-3	C	0.718	0.064	0.892	0.053	0.879	0.045	0.922	0.042	0.946	0.059	0.984	0.035
MFS	DB-1	A	0.870	0.057	0.891	0.051	0.970	0.027	0.964	0.026	0.970	0.058	0.974	0.032
MFS	DB-2	A	0.857	0.042	0.878	0.040	0.957	0.026	0.955	0.038	0.988	0.047	0.999	0.005
MFS	DB-3	A	0.773	0.051	0.817	0.062	0.963	0.020	0.968	0.035	0.995	0.055	0.994	0.016
MFS	DB-1	B	0.834	0.063	0.899	0.062	0.879	0.053	0.887	0.065	0.873	0.066	0.932	0.067
MFS	DB-2	B	0.827	0.051	0.853	0.061	0.896	0.040	0.901	0.060	0.946	0.046	0.997	0.011
MFS	DB-3	B	0.627	0.058	0.776	0.074	0.818	0.040	0.840	0.064	0.832	0.070	0.950	0.064
MFS	DB-1	C	0.882	0.052	0.909	0.050	0.778	0.070	0.862	0.055	0.799	0.064	0.925	0.053
MFS	DB-2	C	0.912	0.037	0.929	0.049	0.893	0.033	0.888	0.053	0.943	0.036	0.993	0.017
MFS	DB-3	C	0.718	0.064	0.811	0.071	0.879	0.045	0.888	0.067	0.946	0.059	0.944	0.090
IFV	DB-1	A	0.870	0.057	0.903	0.046	0.970	0.027	0.968	0.032	0.970	0.058	0.976	0.033
IFV	DB-2	A	0.857	0.042	0.889	0.044	0.957	0.026	0.957	0.038	0.988	0.047	0.999	0.005
IFV	DB-3	A	0.773	0.051	0.880	0.061	0.963	0.020	0.968	0.035	0.995	0.055	0.996	0.010
IFV	DB-1	B	0.834	0.063	0.903	0.048	0.879	0.053	0.910	0.049	0.873	0.066	0.961	0.049
IFV	DB-2	B	0.827	0.051	0.878	0.055	0.896	0.040	0.908	0.059	0.946	0.046	0.996	0.010
IFV	DB-3	B	0.627	0.058	0.883	0.054	0.818	0.040	0.903	0.053	0.832	0.070	0.997	0.037
IFV	DB-1	C	0.882	0.052	0.908	0.047	0.778	0.070	0.892	0.058	0.799	0.064	0.954	0.061
IFV	DB-2	C	0.912	0.037	0.923	0.052	0.893	0.033	0.906	0.048	0.943	0.036	0.993	0.012
IFV	DB-3	C	0.718	0.064	0.887	0.055	0.879	0.045	0.925	0.044	0.946	0.059	0.981	0.024

LBP: Local binary patterns; MFS: Multi-fractal spectrum; IFV: Improved Fisher vectors.

Table 3 Average classification accuracies and standard deviations (±) for specific feature extraction methods, databases and experts

Feature extraction	Data set	Expert	Patch-based				Image-based				Patient-based
			Expert diagnosis		Hybrid diagnosis		Expert diagnosis		Hybrid diagnosis		Expert diagnosis		Hybrid diagnosis
			mean	±	mean	±	mean	±	mean	±	mean	±	mean	±
LBP	mean	mean	0.811	0.091	0.906	0.013	0.893	0.065	0.928	0.026	0.921	0.070	0.978	0.019
MFS	mean	mean	0.811	0.091	0.863	0.052	0.893	0.065	0.906	0.046	0.921	0.070	0.968	0.030
IFV	mean	mean	0.811	0.091	0.895	0.065	0.893	0.065	0.926	0.030	0.921	0.070	0.981	0.017
mean	DB-1	mean	0.862	0.022	0.905	0.007	0.876	0.083	0.917	0.039	0.881	0.074	0.956	0.018
mean	DB-2	mean	0.865	0.037	0.899	0.031	0.915	0.031	0.921	0.027	0.959	0.022	0.996	0.003
mean	DB-3	mean	0.706	0.064	0.859	0.045	0.887	0.063	0.921	0.042	0.924	0.072	0.975	0.020
mean	mean	A	0.833	0.046	0.887	0.006	0.963	0.006	0.963	0.006	0.984	0.011	0.989	0.012
mean	mean	B	0.763	0.102	0.876	0.041	0.864	0.036	0.897	0.023	0.884	0.050	0.968	0.024
mean	mean	C	0.837	0.090	0.900	0.036	0.850	0.054	0.900	0.020	0.896	0.073	0.969	0.025

LBP: Local binary patterns; MFS: Multi-fractal spectrum; IFV: Improved Fisher vectors.