Method and apparatus for computer-aided diagnosis of cancer and product

A non-transitory computer readable storage performing a computer-aided diagnosis of cancer by performing dynamic contrast-enhanced magnetic resonance imaging on a tissue, organ or whole region of a body (ROI) where there might be an abnormality and obtaining a first dataset of images;

• performing a learning step using a second preselected dataset of images with histological confirmance of at least one type of tissue lesion by (i) conducting principal component analysis on the portion of said preselected data correlated with the tissue lesion, (ii) obtaining by principal component analysis a plurality of eigenvectors, (iii) performing analysis of at least the portion of the preselected data by 3TP algorithm labeling to obtain an output of color hue/color intensity coded images with respect to transcapillary transfer constants and extracellular extravascular volume fraction indicative of the at least one type of tissue lesion wherein color intensity reflects changes in signal intensity between the 1st time point (pre-contrast) and the 2nd time point (termed “wash-in rate”), and color hue reflects changes in signal intensity between the 2nd time point and the 3rd time point (termed “wash-out pattern”) according to one preselected color for increased signal intensity, a second preselected color for no significant change, and a third preselected color for decrease in signal intensity, and (iv) manipulating the x and y axes of the eigenvectors around their z-axis to determine a median rotated eigenvector base Erot for at least two eigenvectors, one of which correlates with the wash-in rate and the other of which correlates with wash-out pattern of the 3TP algorithm labeling color hue/color intensity coded images;
• by performing principal component analysis on the first dataset of images according to the algorithm Pd,s=ΓErot−1 wherein Pd,s is the projection of the input data (dataset Γ) on Erot yielding projection coefficient maps of the rotated at least two eigenvectors and enabling detection of a cancerous lesion by a high intensity appearing in the projection coefficient map of the rotated eigenvector correlated with wash-in rate and by a high intensity appearing in the projection coefficient map of the rotated eigenvector correlated with wash-out pattern, and distinguishing between cancerous lesions and benign lesions by each benign lesion appearing in the projection coefficient map of the rotated eigenvector correlated with wash-out pattern showing as a null or negative intensity;
• and by outputting the projection coefficient maps of the rotated at least two eigenvectors.