Diagnosing disease state within image, using Computer-Assisted Diagnosis (CAD) system, comprising: training the system with set of training data images, comprising image features, the image features associated with known classes, and classes associated with predetermined possible clinical actions； determining cost function of weighted error terms, and parameters, where certain parameters for certain image features are weighted due to known clinical significance in diagnosis； receiving image, comprising image features； using system to give specific clinical action by extracting image feature and applying weighted cost function to identify class. Diagnosing disease state within image, using Computer-Assisted Diagnosis system, comprising: acquiring image； identifying region of interest (ROI) within image； using CAD system, configured to recommend clinical action based on minimisation of discrepancies between evidence based action and recommended action by another system for specific users. Clinical action may provide diagnostic category associated with biopsy. Weighted error terms may correspond to errors from: specific operators； institution； locale； workflow position； aggregation of errors made by number of operators. Classifier may comprise: neural networks； support vector machines； naïve Bayes classifier； genetic programming； reinforcement learning； deep neural networks. Embodiment: diagnosing breast cancer, using mammograms, with Breast Imaging Reporting and Data System (BI-RADS) categories, avoiding unnecessary biopsies.
