Back

OCR-Mediated Modality Dominance in Vision-Language Models: Implications for Radiology AI Trustworthiness

2026-02-24 health informatics Title + abstract only
View on medRxiv
Show abstract

1.BackgroundVision-language models (VLMs) are increasingly proposed for radiologic decision support, yet the security implications of deploying general-domain, OCR-capable models in diagnostic workflows remain poorly characterized. When image-embedded text is not treated as untrusted input, the visual channel becomes vulnerable to adversarial manipulation through OCR-readable overlays. MethodsNine commercial VLMs, none intended or validated for clinical diagnosis, were evaluated on 600 brain MR...

Predicted journal destinations

1
npj Digital Medicine
85 training papers
Top 0.3% 17.4%
2
Journal of the American Medical Informatics Association
53 training papers
Top 2% 11.6%
3
PLOS Digital Health
88 training papers
Top 1% 8.8%
4
Scientific Reports
701 training papers
Top 40% 6.3%
5
JAMIA Open
35 training papers
Top 3% 5.8%
6
BMC Medical Informatics and Decision Making
36 training papers
Top 2% 5.8%
7
Journal of Biomedical Informatics
37 training papers
Top 2% 5.8%
8
Nature Communications
483 training papers
Top 19% 5.8%
9
Computers in Biology and Medicine
39 training papers
Top 2% 3.1%
10
PLOS ONE
1737 training papers
Top 89% 3.0%
11
International Journal of Medical Informatics
25 training papers
Top 3% 1.9%
12
JMIR Medical Informatics
16 training papers
Top 3% 1.9%
13
Journal of Medical Internet Research
81 training papers
Top 9% 1.9%
14
Scientific Data
30 training papers
Top 2% 1.3%
15
JAMA Network Open
125 training papers
Top 17% 1.3%
16
BMC Medical Research Methodology
41 training papers
Top 4% 1.3%
17
eBioMedicine
82 training papers
Top 7% 1.2%
18
Patterns
15 training papers
Top 2% 0.9%
19
Cureus
64 training papers
Top 17% 0.9%
20
Nature Medicine
88 training papers
Top 13% 0.9%
21
PLOS Computational Biology
141 training papers
Top 15% 0.7%
22
Frontiers in Digital Health
18 training papers
Top 4% 0.7%
23
Frontiers in Oncology
34 training papers
Top 6% 0.7%
24
The Lancet Digital Health
25 training papers
Top 2% 0.7%
25
Communications Medicine
63 training papers
Top 8% 0.7%