PROJECT TITLE :

Development and Assessment of a New Global Mammographic Image Feature Analysis Scheme to Predict Likelihood of Malignant Cases

ABSTRACT:

Researchers in this project hope to create and test a new Computer-Aided Diagnosis (CADx) method based on a comprehensive analysis of global mammographic imaging features to determine the likelihood that a given instance is malignant. A retrospective photographic collection of 1,959 cases was created. In each case, suspicious lesions were found and biopsied. There are 737 cancerous instances and 1,222 noncancerous. Four mammograms of the left and right breasts, taken from the craniocaudal and mediolateral oblique perspectives, are included in each case. Applying the discrete cosine transform (DCT) and fast Fourier transform (FFT) to build two image maps, compute bilateral image feature differences from left and right breasts, use a support vector machine (SVM) for the prediction of malignantness.. The original mammograms and two transformation maps were used to generate three subsets of image features. To train and test the models, a 10-fold cross-validation method was used for each of the four models. Image characteristics from one of three sub-groups were used to construct AUCs ranging in value from 0.85 to 0.91, respectively. AUC = 0.960.01 (p; 0.01) is substantially higher with AUC = 0.960.01 (p; 0.01) when all image features computed in three groups are combined. This work shows that a new global image feature analysis-based CADx system for high-performance mammograms may be developed. This novel CADx technique is more efficient in development and perhaps more robust in future use since it avoids difficulties and possible errors in breast lesion segmentation.