Application of V-detector in dental diagnosis To be submitted to CEC 2006.

Application of V-detector Application of V-detector in dental diagnosisin dental diagnosis

To be submitted to CEC 2006To be submitted to CEC 2006

backgroundbackground

Malocclusion – diagnosis using X-rayMalocclusion – diagnosis using X-ray

V-detector – one-class classificationV-detector – one-class classification

malocclusionmalocclusion

Different types: I (normal bite), II Different types: I (normal bite), II (overbite), and III (underbite)(overbite), and III (underbite)Mild or severe (functional)Mild or severe (functional)

Lateral view skull X-rayLateral view skull X-ray

Existing diagnosis methodExisting diagnosis methodAngle’s classification: angle ANB Angle’s classification: angle ANB (3 in the picture)(3 in the picture)

N

A

B

Feature extractionFeature extraction

Brightness distribution instead of entity Brightness distribution instead of entity identificationidentification

Binarization at multiple thresholdBinarization at multiple threshold

Quantitatize each binary image with four Quantitatize each binary image with four real numbersreal numbers

Remove artificial partsRemove artificial parts

Binarization using multiple Binarization using multiple thresholdsthresholds

Choose thresholds & decide reference pointChoose thresholds & decide reference point

T0 = Vmax,

T1 = Vmax − (Vmax − Vmin)/nT ,

...,

TnT−1 = Vmax − (nT − 1)(Vmax − Vmin)/nT ,

Binarized at the highest threshold

Extract four featuresExtract four featuresat each thresholdat each threshold

(a) Horizontal displacement

x = xwhite − x0,

(b) Vertical displacement

y = ywhite − y0,

(c) Displacement distancer = mean of distances between white pixels to (x0, y0)

(d) Area mass

A = total number of white pixel/width · height

Experiment resultsExperiment results

Compare with SVMCompare with SVM

Using half of normal data to trainUsing half of normal data to train

summarysummary

A novel feature extraction is proposed.A novel feature extraction is proposed.

V-detector shows some potentials.V-detector shows some potentials.

Issue: a lot more normal data are desired.Issue: a lot more normal data are desired.