Prompt2Perturb (P2P): Text-Guided Diffusion-Based Adversarial Attacks on\n Breast Ultrasound Images
Bibliographic record
Abstract
Deep neural networks (DNNs) offer significant promise for improving breast\ncancer diagnosis in medical imaging. However, these models are highly\nsusceptible to adversarial attacks--small, imperceptible changes that can\nmislead classifiers--raising critical concerns about their reliability and\nsecurity. Traditional attacks rely on fixed-norm perturbations, misaligning\nwith human perception. In contrast, diffusion-based attacks require pre-trained\nmodels, demanding substantial data when these models are unavailable, limiting\npractical use in data-scarce scenarios. In medical imaging, however, this is\noften unfeasible due to the limited availability of datasets. Building on\nrecent advancements in learnable prompts, we propose Prompt2Perturb (P2P), a\nnovel language-guided attack method capable of generating meaningful attack\nexamples driven by text instructions. During the prompt learning phase, our\napproach leverages learnable prompts within the text encoder to create subtle,\nyet impactful, perturbations that remain imperceptible while guiding the model\ntowards targeted outcomes. In contrast to current prompt learning-based\napproaches, our P2P stands out by directly updating text embeddings, avoiding\nthe need for retraining diffusion models. Further, we leverage the finding that\noptimizing only the early reverse diffusion steps boosts efficiency while\nensuring that the generated adversarial examples incorporate subtle noise, thus\npreserving ultrasound image quality without introducing noticeable artifacts.\nWe show that our method outperforms state-of-the-art attack techniques across\nthree breast ultrasound datasets in FID and LPIPS. Moreover, the generated\nimages are both more natural in appearance and more effective compared to\nexisting adversarial attacks. Our code will be publicly available\nhttps://github.com/yasamin-med/P2P.\n
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 | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.002 | 0.003 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.003 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.007 | 0.008 |
| Research integrity | 0.002 | 0.007 |
| Insufficient payload (model declined to judge) | 0.001 | 0.003 |
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; both teacher heads agree on what is shown here.
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".