Biomedical Engineering Reference
In-Depth Information
Finally, the classification results of breast tumors for the noncontinuous and
continuous strain imaging methods are given in Table 2. In order to estimate
the performance of the proposed method, five items are compared: accuracy,
sensitivity, specificity, positive predictive value (PPV), and negative predictive
value (NPV) (see Table 3). According to Tables 2 and 3, continuous strain analysis
yields more stable and better results than conventional (noncontinuous) strain
analysis, in which only two images were used.
5.5. Discussion on 2D Strain Data Analysis
In this study the level set segmentation approach was used to extract tumor
contours from continuous strain images. Because the 2D method was inadequate
for calculating the volume of the sideslip during compression, the 3D level set
method was used. The results of the automated segmentation were found to be
acceptable in all cases with a radiologist's review. According to these satisfying
results, we can confirm the feasibility of this proposed framework. However, the
accuracy of the CAD system is strongly dependent on an exact segmentation result.
In addition to adopting some image preprocessing techniques before using the level
set approach, we need to find more robust methods to improve the segmentation
result.
The values of all proposed features showed that malignant tumors tend to be
more rigid and have more desmoplastic reactions within the surrounding tissue,
which means that those features of strain change to a lesser extent on continuous
ultrasound images. From the experimental results given in Table 3, the accuracy
of continuous strain imaging was 87.00% (87/100), sensitivity 85.00% (34/40),
specificity 88.33% (53/60), PPV 82.93% (36/41), and NPV 89.83% (53/59). We
can conclude that the proposed method with these strain features is better than
conventional noncontinuous strain analysis.
There were limitations in this study. In clinical applications of ultrasound
imaging of the strain on tissue from probe compression, the operator must apply
constant pressure on the scanning probe to prevent an inclined scanning plane.
Otherwise, the tumor may slip out of the scanning plane, which will create inaccu-
racies. Gradual compressionwith the ultrasound probe, however, can be performed
easily after some practice or by using a compression plate. Compared to the accu-
racy of elastography with radiofrequency data, the accuracy of our method heavily
depends on the results of lesion segmentation. We adjusted some parameters to
control the segmented contours and to obtain better results with an expert radi-
ologist's input. Currently, the computerized segmentation in our method is only
semi-automated. For tumors with poorly demarcated borders, a good preprocess-
ing technique would be helpful to enhance tumor boundaries. For tumors without
a definable boundary or for isoechoic masses, other methods should be developed.
For example, some representative reference points could be used to evaluate tissue
Search WWH ::




Custom Search