ICAM: Interpretable Classification via Disentangled Representations and Feature Attribution Mapping
Authors: Cher Bass, Mariana da Silva, Carole Sudre, Petru-Daniel Tudosiu, Stephen Smith, Emma Robinson
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our method on 2D and 3D brain image datasets of dementia (ADNI dataset), ageing (UK Biobank), and (simulated) lesion detection. We show that FA maps generated by our method outperform baseline FA methods when validated against ground truth. We evaluate the performance of ICAM through studies on three datasets to perform 1) ablation studies on 2D simulations; 2) evaluation of the accuracy of the generated attribution maps (using ground truth disease conversion maps derived from the ADNI dataset); 3) exploration of the flexibility of the approach for investigating phenotypic variation (using healthy ageing data from UK Biobank). |
| Researcher Affiliation | Academia | Cher Bass Kings College London London, UK, WC2R 2LS cher.bass@kcl.ac.uk Mariana da Silva Kings College London London, UK, WC2R 2LS Carole Sudre Kings College London London, UK, WC2R 2LS Petru-Daniel Tudosiu Kings College London London, UK, WC2R 2LS Stephen M. Smith University of Oxford Oxford, UK, OX1 2JD Emma C. Robinson Kings College London London, UK, WC2R 2LS emma.robinson@kcl.ac.uk |
| Pseudocode | No | The paper describes the architecture and loss functions but does not provide any pseudocode or algorithm blocks. |
| Open Source Code | Yes | In addition, our code, which has been released on Git Hub at https://github. com/Cher Bass/ICAM, extends to multi-class classification and regression tasks. |
| Open Datasets | Yes | We validate our method on 2D and 3D brain image datasets of dementia (ADNI dataset), ageing (UK Biobank), and (simulated) lesion detection. We used the HCP dataset [16, 45], with T2 MRI data, for simulating lesions in MRI brain slices. We use the longitudinal ADNI dataset [23], with T1 MRI data. We used T1 MRI data from the UK Biobank [1, 32]. |
| Dataset Splits | No | The paper states: "For a fair comparison, we use the same training, validation and testing datasets." and "We used these subjects in our validation and test sets". However, it does not provide specific percentages or counts for these splits in the main text needed to reproduce the data partitioning. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types) used to run its experiments. |
| Software Dependencies | No | The paper does not provide specific version numbers for any ancillary software dependencies used in the experiments. |
| Experiment Setup | No | The paper briefly describes network architecture components but does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs, optimizer settings) or detailed system-level training configurations in the main text. It mentions lambda values are in supplementary section A.6, but the question asks for details in the main text. |