Toward Improved Interpretability in Medical Imaging: Revealing the Disease Evidence From Chest X-Ray Images Using an Adversarial Generative Approach
Bibliographic record
Abstract
In recent years, deep neural networks have made significant progress in the field of automated disease recognition for medical imaging. Although methods based on deep neural networks reported good results in automated disease classification, many of them lack output interpretability. Unlike methods that only aim at predicting if an input image contains disease or not, our work focuses additionally on revealing the disease evidence and offering interpretability of decision-making. We assume that an abnormal image is the additive composition of a hypothetical but synthesizable normal counterpart and a related disease effect map. Therefore we propose a disease decomposition network (DDN) that can explicitly locate, separate the disease evidence from an abnormal image and synthesize this hypothetical normal counterpart simultaneously. The proposed DDN has two branches: one is for translating the input image to a normal image, and the other is for locating the underlying abnormalities accurately if the input is abnormal. The novelty of our proposed DDN is that it not only provides highly accurate visual disease evidence maps, but also generates corresponding healthy images that are close to real normal chest X-rays (CXRs). From an interpretability and user interface perspective, it is insightful and helpful to provide a relevant reference (albeit hypothetical) for better understanding the difference between a normal and diseased CXR image. We evaluated our method on an extensive and well-established public chest X-ray dataset. The results demonstrate that the proposed DDN can translate abnormal CXR images to healthy CXRs while also generating disease evidence maps with good localization performance over other state-of-the-art methods.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".