Program Information

Image Analysis in Ultrasonography for Diagnosis of Sjoegren's Syndrome Using Dual-Tree Complex Wavelet Transform

T Matsui¹*, M Ohki² , T Nakamura³ , Y Takagi³ , (1) Department of Health Sciences, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan, (2) Department of Health Sciences, Faculty of Medical Sciences, Kyushu University, Fukuoka, Japan, (3) Department of Radiology and Cancer Biology, Nagasaki University School of Dentistry, Nagasaki, Japan

Presentations

SU-E-I-30 Sunday 3:00PM - 6:00PM Room: Exhibit Hall

Purpose:
Sjoegren's syndrome (SS) is an autoimmune disease invading mainly salivary and lacrimal glands. Ultrasonography is used for an initial and non-invasive examination of this disease. However, the ultrasonography diagnosis tends to lack in objectivity and depends on the operator's skills. The purpose of this study is to propose a computer-aided diagnosis (CAD) system for SS based on a dual-tree complex wavelet transform (DT-CWT) and machine learning.

Methods:
The subjects of this study were 174 patients suspected of having SS at Nagasaki University Hospital and examined with ultrasonography of the parotid glands. Out of these patients, 77 patients were diagnosed with SS by sialography. A region of interest (ROI) of 128 x 128 pixels was set within the parotid gland that was indicated by a dental radiologist. The DT-CWT was applied to the images in the ROI and every image was decomposed into 72 sub-images of the real and imaginary components in six different resolution levels and six orientations. The statistical features of the sub-image were calculated and used as data input for the support vector machine (SVM) classifier for the detection of SS. A ten-fold cross-validation was employed to verify the result of SVM. The accuracy of diagnosis was compared by a CAD system with a human observer performance.

Results:
The sensitivity, specificity, and accuracy in the detection of SS were 95%, 86%, and 91% through our CAD system respectively, while those by a human observer were 84%, 81%, and 83% respectively.

Conclusion:
The proposed computer-aided diagnosis system for Sjoegren's syndrome in ultrasonography based on dual-tree complex wavelet transform had a better performance than a human observer.

Contact Email: