Abstract:
HTML
Purpose: Computer-aided diagnostic (CAD) scheme for lung nodule detection in
chest radiographs has been developed as a tool for lung cancer screening.
However, it is difficult to obtain a low false-positive rate because there are
a large variety of normal structures similar to nodules in chest radiographs.
Our purpose in this study is to develop a new technique for false positive
reduction in a CAD scheme for lung nodule detection in chest radiographs by use
of a multiple massive training artificial neural network (Multi-MTANN).
Methods and Materials: We developed a Multi-MTANN that consists of nine massive
training artificial neural networks (MTANNs) arranged in parallel. The MTANN is
a highly nonlinear filter that performs both enhancement of nodules and
suppression of a specific type of non-nodule. The inputs of the MTANN were the
pixel values in a sub-region extracted from a chest radiograph. The teacher
image for the MTANN was designed to contain the distribution for the
"likelihood of being a nodule." Each MTANN was trained by using
typical nodules and non-nodules representing a specific type of non-nodule. For
example, MTANN No. 1 was trained to distinguish nodules from ribs with high
contrast; MTANN No. 2, No. 3, and No. 4 were applied to eliminate ribs
overlapping with vessels; ribs overlapping with soft-tissue; parts of normal
structures; and so on. Nine MTANNs were combined by using the logical AND
operation such that each of the trained MTANNs removed the specific type of
non-nodule, and thus various types of non-nodules can be eliminated. Our
database consisted of 91 solitary pulmonary nodules including 64 malignant
nodules and 27 benign nodules in 91 chest radiographs. All nodules were
confirmed by CT examinations, and the average size of the nodules was 24.9 mm.
Results: With our current CAD scheme based on a difference image technique
together with rule-based and linear discriminant classifiers, a detection
sensitivity of 82.4% (75/91 nodules) with 410 false positives (4.5 per image)
was achieved. The trained Multi-MTANN was applied for further reduction of the
false positives. The results indicated that 55.1% (226/410) of the false
positives were removed with a reduction of one true positive. The
false-positive rate of our current CAD scheme was improved from 4.5 to 2.0
false positives per image at the overall sensitivity of 81.3% (74/91 nodules).
Conclusion: By use of the Multi-MTANN, the false positive rate of a CAD scheme
for lung nodule detection in chest radiographs can be improved substantially,
while the current sensitivity is maintained. (K.D. is a shareholder in R2
Technology, Inc. and Deus Technology, Inc.)
Questions about this event email: suzuki@uchicago.edu
Suzuki PhD, K,
False Positive Reduction in Computerized Detection of Lung Nodules in Chest Radiographs Using Massive Training Artificial Neural Network. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3102983.html