RATIONALE AND OBJECTIVES: The aim of this study was to develop a semi-automated computer-aided diagnosis (CAD) system based on high-resolution ultrasonography for classifying benign and malignant soft tissue tumors (STTs). MATERIALS AND METHODS: One hundred seven patients with STTs (70 benign and 37 malignant) were enrolled, and regions of interest were manually delineated for analysis. Sixteen tumor shape features, including five geometric features and 11 morphologic features (six old and five new normalized radial length [NRL] features) were individually evaluated using Student's t test and the area under the receiver-operating characteristic curve (A(z)). Then linear discriminant analysis with stepwise feature selection was used to construct a semi-automated CAD system with old NRL features, new NRL features, and all features combined. Additionally, two experienced radiologists participated in malignancy grading of tumors. To investigate the associations among CAD results, pathologic results, and radiologists' rankings, Spearman's rank correlation coefficient was used in the statistical analysis. RESULTS: The results showed that 11 features had P values < .05, and five of the proposed features were significant. The optimal CAD system achieved accuracy of 87.9%, sensitivity of 89.2%, specificity of 87.1%, and an A(z) value of 0.93. Correlation between pathologic results and radiologists' rankings was obtained (radiologist A: r=0.62, P < .01; radiologist B: r=0.61, P < .01). In addition, a higher correlation between pathologic results and CAD results (r=0.73, P < .01) was demonstrated. CONCLUSION: This semi-automated CAD method based on tumor shape features can successfully distinguish between benign and malignant STTs. It can also provide a second opinion to ultrasound for the diagnosis of STTs.
RATIONALE AND OBJECTIVES: The aim of this study was to develop a semi-automated computer-aided diagnosis (CAD) system based on high-resolution ultrasonography for classifying benign and malignant soft tissue tumors (STTs). MATERIALS AND METHODS: One hundred seven patients with STTs (70 benign and 37 malignant) were enrolled, and regions of interest were manually delineated for analysis. Sixteen tumor shape features, including five geometric features and 11 morphologic features (six old and five new normalized radial length [NRL] features) were individually evaluated using Student's t test and the area under the receiver-operating characteristic curve (A(z)). Then linear discriminant analysis with stepwise feature selection was used to construct a semi-automated CAD system with old NRL features, new NRL features, and all features combined. Additionally, two experienced radiologists participated in malignancy grading of tumors. To investigate the associations among CAD results, pathologic results, and radiologists' rankings, Spearman's rank correlation coefficient was used in the statistical analysis. RESULTS: The results showed that 11 features had P values < .05, and five of the proposed features were significant. The optimal CAD system achieved accuracy of 87.9%, sensitivity of 89.2%, specificity of 87.1%, and an A(z) value of 0.93. Correlation between pathologic results and radiologists' rankings was obtained (radiologist A: r=0.62, P < .01; radiologist B: r=0.61, P < .01). In addition, a higher correlation between pathologic results and CAD results (r=0.73, P < .01) was demonstrated. CONCLUSION: This semi-automated CAD method based on tumor shape features can successfully distinguish between benign and malignant STTs. It can also provide a second opinion to ultrasound for the diagnosis of STTs.