| Description |
viii, 68 leaves : illustrations (some colored) ; 29 cm |
| Summary |
"This thesis demonstrates an algorithm to identify the features of radial streaming and pseudopods found in dermoscopy images of malignant melanoma. The algorithm includes preprocessing images using the Principal Component Transform and an application of mathematical techniques to extract the desired features from the lesion image on the basis of color, size, shape, and orientation with respect to the lesion boundary. These identified features are further used to classify the lesion as malignant or benign. The proposed algorithm has produced a diagnostic accuracy of 90.50% for the training set of 31 malignant and 109 benign images and a diagnostic accuracy of 93.52% for the test set of 31 malignant and 110 benign images using Receiver Operating Characteristic curve analysis. The code is implemented in MATLABĀ® 7.4.0.287 (R2007a)"--Abstract, leaf iii. |
|