Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2508.01016

Diagnostic Accuracy of Open-Source Vision-Language Models on Diverse Medical Imaging Tasks

Gustav Müller-Franzes·Debora Jutz·Jakob Nikolas Kather·Christiane Kuhl·Sven Nebelung·Daniel Truhn
View original source ↗Read source document ↗Export metadata ↓

Abstract

This retrospective study evaluated five VLMs (Qwen2.5, Phi-4, Gemma3, Llama3.2, and Mistral3.1) using the MedFMC dataset. This dataset includes 22,349 images from 7,461 patients encompassing chest radiography (19 disease multi-label classifications), colon pathology (tumor detection), endoscopy (colorectal lesion identification), neonatal jaundice assessment (skin color-based treatment necessity), and retinal fundoscopy (5-point diabetic retinopathy grading). Diagnostic accuracy was compared in three experimental settings: visual input only, multimodal input, and chain-of-thought reasoning. Model accuracy was assessed against ground truth labels, with statistical comparisons using bootstrapped confidence intervals (p<.05). Qwen2.5 achieved the highest accuracy for chest radiographs (90.4%) and endoscopy images (84.2%), significantly outperforming the other models (p<.001). In colon pathology, Qwen2.5 (69.0%) and Phi-4 (69.6%) performed comparably (p=.41), both significantly exceeding other VLMs (p<.001). Similarly, for neonatal jaundice assessment, Qwen2.5 (58.3%) and Phi-4 (58.1%) showed comparable leading accuracies (p=.93) significantly exceeding their counterparts (p<.001). All models struggled with retinal fundoscopy; Qwen2.5 and Gemma3 achieved the highest, albeit modest, accuracies at 18.6% (comparable, p=.99), significantly better than other tested models (p<.001). Unexpectedly, multimodal input reduced accuracy for some models and modalities, and chain-of-thought reasoning prompts also failed to improve accuracy. The open-source VLMs demonstrated promising diagnostic capabilities, particularly in chest radiograph interpretation. However, performance in complex domains such as retinal fundoscopy was limited, underscoring the need for further development and domain-specific adaptation before widespread clinical application.

Explore related subjects

eess.IVcs.CV

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Gustav Müller-Franzes, Debora Jutz, Jakob Nikolas Kather, Christiane Kuhl, Sven Nebelung, Daniel Truhn. 2025-08-01. Diagnostic Accuracy of Open-Source Vision-Language Models on Diverse Medical Imaging Tasks. https://arxiv.org/abs/2508.01016

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

VideoPulse: Neonatal heart rate and peripheral capillary oxygen saturation (SpO2) estimation from contact free video

Remote photoplethysmography (rPPG) enables contact free monitoring of vital signs and is especially valuable for neonates, since conventional methods often require sustained skin contact with adhesive probes that can irritate fragile skin and increase infection control burden. We present VideoPulse, a neonatal dataset and an end to end pipeline that estimates neonatal heart rate and peripheral capillary oxygen saturation (SpO2) from facial video. VideoPulse contains 157 recordings totaling 2.6 hours from 52 neonates with diverse face orientations. Our pipeline performs face alignment and artifact aware supervision using denoised pulse oximeter signals, then applies 3D CNN backbones for heart rate and SpO2 regression with label distribution smoothing and weighted regression for SpO2. Predictions are produced in 2 second windows. On the NBHR neonatal dataset, we obtain heart rate MAE 2.97 bpm using 2 second windows (2.80 bpm at 6 second windows) and SpO2 MAE 1.69 percent. Under cross dataset evaluation, the NBHR trained heart rate model attains 5.34 bpm MAE on VideoPulse, and fine tuning an NBHR pretrained SpO2 model on VideoPulse yields MAE 1.68 percent. These results indicate that short unaligned neonatal video segments can support accurate heart rate and SpO2 estimation, enabling low cost non invasive monitoring in neonatal intensive care.

eess.IV↗
arXiv2026

Synthetic Fingerprints for Children Under Four: Generation and Biometric Evaluation

Fingerprint recognition in children under four is of interest for longitudinal identity applications, but research in this age range is constrained by the limited availability and sensitivity of real fingerprint data. Synthetic data may provide a useful complementary resource if generated samples are carefully evaluated for biometric quality, similarity to the real training data, and identity diversity. This paper presents an evaluation and selection protocol for synthetic fingerprints generated from a small pediatric dataset. The protocol was applied to 32,000 candidates produced by an age-conditioned generator fine-tuned with 205 fingerprints from nine children. Candidates were evaluated using NFIQ 2, NBIS minutiae extrac-tion and matching, fingerprint-pattern classification, similarity to the complete real reference set, and pairwise similarity among retained synthetic samples. After candidate filtering and a final symmetric pairwise verification, 1,985 synthetic fingerprints remained. The youngest age group continued to produce retained samples within the sampling budget, whereas the oldest original age group produced 24 retained prints. A data-derived two-group age representation improved pattern-distribution agreement for the younger group. Repeated renderings of fixed synthetic iden-tities produced minutiae-based mated scores comparable to the real mated scores, although a DINOv2 texture representation showed substantially greater within-generator similarity. The results indicate that synthetic pediatric fingerprints can support controlled research use, but their evaluation should include both real-to-synthetic similarity and synthetic-to-synthetic diversity, and conclusions should remain specific to the matchers and measurements used.

eess.IV↗
arXiv2026

Adaptive Color Grading

Independent control of tonescale regions (e.g., shadows, highlights) is essential for painters, photographers and cinematographers to bring 2D images to life. In image manipulation software this is most directly addressed by color grading modules, which use intensity thresholds to segment distinct illumination regions for local manipulation. In this work we develop an open source color grading tool and use it to annotate a large dataset of video frames with tonescale region thresholds. Using these thresholds we conduct modeling experiments with strategies based on both practitioners' conventional wisdom and machine learning. Results show that K-nearest neighbors is an effective prediction strategy, outperforming state-of-the-art end-to-end methods for image enhancement. This outcome demonstrates the benefit of focusing on a compact set of core parameters when modeling creative stylization processes. Our adaptive color grading interface and data are available at https://github.com/SamsungLabs/adaptive-color-grading.

eess.IV↗