Abstract:
HTML
Purpose: To develop an advanced computer-aided diagnostic (CAD) scheme for
lung cancer diagnosis in chest radiographs and evaluate the scheme with an
observer performance study.
Methods and Materials: The advanced CAD scheme consisted of two major
components, one for detection of lung nodules, and another for distinguishing
lung cancers from computer false positives and benign nodules. The CAD output
was indicated by color-coded arrows which indicated both computer detection
results of suspicious nodules, and computer-estimated likelihood of malignancy.
In addition, the CAD scheme allowed to provide computer-estimated likelihood of
malignancy for potential lesions interactively, if the radiologist indicated
any suspected areas on chest radiographs. Receiver operating characteristic
(ROC) analysis for detecting lung cancers was performed without and with CAD
output; radiologists were asked to indicate the likelihood of malignancy for
the case, as well as the location of most suspicious area for lung cancer. Our
database included 150 chest radiographs with 72 confirmed solitary lung
cancers, 36 confirmed solitary benign nodules, and 42 normal cases, which were
used for evaluation of the CAD scheme. For the observer performance study, 48
of these chest radiographs were used, with the same balance of cancer and other
cases. Twelve radiologists participated in this ROC study. Monochrome and color
LCD monitors, each with an active area of 408 mm x 306 mm (1600 x 1200 pixels),
were used for displaying the original image and the image with CAD output,
respectively.
Results: The CAD scheme achieved a performance of 81.4 % sensitivity with 1.2
false positives per image for detection of lung nodules, and an area under the
ROC curve (Az) of 0.863 for distinction of lung cancers from other
computer-detected false positives and benign nodules. The average Az values for
radiologists without and with CAD output were 0.800 and 0.753, respectively.
The performance of radiologists was improved significantly when this advanced
CAD scheme was used (P = 0.008). LROC curves for all observers with CAD output
were also improved from those obtained without CAD output.
Conclusion: This advanced CAD scheme including detection and classification of
lung cancers would be useful in improving the performance of radiologists for
detecting lung cancers on chest radiographs. (K.D. is a shareholder in R2
Technology, Inc. and Deus Technology, Inc. This work is supported by USPHS
Grant CA62625.)
Questions about this event email: junji@uchicago.edu
Shiraishi PhD, J,
Usefulness of Advanced Computer-aided Diagnostic Scheme for Lung Cancer Diagnosis in Chest Radiographs: Observer Performance Study. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3101054.html