Abstract:
HTML
Purpose: It is difficult for radiologists to make correct clinical decisions
for biopsy or follow-up on clustered microcalcifications on mammograms by
taking into account possible histological classifications. The purpose of this
study is to develop an automated computerized scheme for identifying
histological classification of clustered microcalcifications on mammograms in
order to assist radiologists' interpretation as a "second opinion".
Methods and Materials: Our database consisted of 58 magnification mammograms
(512x512pixels, 12bit/pixels, 0.025mm/pixel), which included 35 malignant
clustered microcalcifications (invasive carcinoma, noninvasive carcinoma of
comedo type, and noninvasive carcinoma of noncomedo type) and 23 benign
clustered microcalcifications (mastopathy and fibroadenoma). The histological
classification of all clusters were proved by pathological diagnosis. The
clustered microcalcifications were first segmented by using a filter bank and a
thresholding technique. Five objective features on clustered
microcalcifications were determined by taking into account subjective features
with which radiologists commonly use to estimate the likelihood of malignancy.
These features were: (1)the variation in the size of microcalcifications within
a cluster, (2)the variation in pixel values of microcalcifications within a
cluster, (3)the irregularity in the shape of microcalcifications within a
cluster, (4)the extent of linear and branching distribution of
microcalcifications, and (5)the distribution of microcalcifications in the
direction toward nipple. Bayes decision rule with five features was employed
for distinguishing between five histological classifications.
Results: The sensitivity and the specificity of this computerized scheme for
distinction between malignant and benign clustered microcalcifications were
97.1%(34/35) and 95.7%(22/23), respectively. The sensitivity for distinguishing
between three abnormal histological classifications was 77.8%(7/9) for invasive
carcinoma, 75.0%(9/12) for noninvasive carcinoma of comedo type, 92.9%(13/14)
for noninvasive carcinoma of noncomedo type. The specificity for distinguishing
between two benign histological classifications was 94.1%(16/17) for
mastopathy, and 100.0%(6/6) for fibroadenoma.
Conclusion: This automated computerized scheme may be useful to assist
radiologists in their assessment of clustered microcalcification. (S.K, K.D.
are shareholders in R2 Technology. K.D is shareholder in Deus Technologies.)
Questions about this event email: nakayama@clin.medic.mie-u.ac.jp
Nakayama MS, R,
Computer-aided Diagnosis Scheme for Histological Classification of Clustered Microcalcifications on Magnification Mammograms. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3103923.html