We propose a CNN based technique that aggregates feature maps from its\nmultiple layers that can localize abnormalities with greater details as well as\npredict pathology under consideration. Existing class activation mapping (CAM)\ntechniques extract feature maps from either the final layer or a single\nintermediate layer to create the discriminative maps and then interpolate to\nupsample to the original image resolution. In this case, the subject specific\nlocalization is coarse and is unable to capture subtle abnormalities. To\nmitigate this, our method builds a novel CNN based discriminative localization\nmodel that we call high resolution CAM (HR-CAM), which accounts for layers from\neach resolution, therefore facilitating a comprehensive map that can delineate\nthe pathology for each subject by combining low-level, intermediate as well as\nhigh-level features from the CNN. Moreover, our model directly provides the\ndiscriminative map in the resolution of the original image facilitating finer\ndelineation of abnormalities. We demonstrate the working of our model on a\nsimulated abnormalities data where we illustrate how the model captures finer\ndetails in the final discriminative maps as compared to current techniques. We\nthen apply this technique: (1) to classify ependymomas from grade IV\nglioblastoma on T1-weighted contrast enhanced (T1-CE) MRI and (2) to predict\nParkinson's disease from neuromelanin sensitive MRI. In all these cases we\ndemonstrate that our model not only predicts pathologies with high accuracies,\nbut also creates clinically interpretable subject specific high resolution\ndiscriminative localizations. Overall, the technique can be generalized to any\nCNN and carries high relevance in a clinical setting.\n