Automatic detection of anomalies in screening mammograms

BMC Medical Imaging

Table 2 Comparison of the performance of wavelet bases on the DDSM dataset

Wavelet basis^§	Best feature combination			Sensitivity^*(%)	Specificity^†(%)	Classification rate^‡(%)
Haar	M-h1	M-d1	S-h3	99.2	36.6	60.3
Db 2	M-h3	M-d8	S-h5	97.4	42.7	63.4
Db 4	M-h8	M-d1	S-h5	95.2	20.8	49
Db 8	M-h6	S-v8	S-d3	97.5	40.4	62
Bior 1.5	M-d4	S-h6	---	96.9	38.8	60.8
Bior 2.2	M-h5	M-v2	S-d2	98.8	44.8	65.2
Bior 2.8	M-d4	S-d2	S-a5	92.9	46.9	64.4
Bior 3.7	M-d4	S-h4	S-d4	98.9	28.1	54.9
Bior 4.4	M-h1	M-d4	S-d2	96.1	43	63.1
Bior 5.5	M-h6	M-d5	S-d2	98.5	38.1	61
Bior 6.8	M-v3	M-d4	S-d2	98	39	61.3

§Wavelet basis notation: Dbn where n describes the number of coefficients used in the wavelet. Db2 encodes polynomials with two coefficients, i.e. constant and linear components. Biorm.n where n describes the order for decomposition and m is the order used for reconstruction.
*Sensitivity is defined as TP/(TP + FN).
†Specificity is defined as TN/(TN + FP).
‡Classification rate is defined as (TP + TN)/(TP + TN + FP + FN).
The best triplet feature was selected for each wavelet.

ISSN: 1471-2342