Pocket-Dentist: On-Device Dental Image Understanding via Efficient Multimodal Large Language Models

2026-05-27

AI Summary

This study addresses the lack of standardized evaluation and neglect of computational efficiency in existing dental vision-language models, which hinders their deployment in resource-constrained, privacy-sensitive settings. To bridge this gap, we introduce an efficiency-oriented multimodal dental question-answering benchmark encompassing three datasets, five task categories, and seven evaluation metrics, and systematically assess the trade-offs between accuracy and computational cost across 14 models. Our analysis reveals that smaller-scale models (e.g., 2B parameters) outperform larger counterparts in dental image understanding, achieving higher accuracy with significantly lower computational overhead. Building on this insight, we propose Pocket-Dentist-2B, a lightweight multimodal large language model enhanced with instruction tuning and mobile-optimized inference, enabling on-device deployment on an iPhone 17 Pro with a per-sample latency of only 4.31 seconds—4.9× faster and 2.3× more memory-efficient than a 7B baseline.

Abstract

Evaluations of dental vision-language models remain fragmented across datasets, task definitions and metrics, and often ignore their computational cost. This limits their widespread deployment for dental screening outside specialist centres, where timely inference, limited hardware, and local handling of patient images are vital for practical, privacy-preserving clinical prescreening. Here we present Pocket-Dentist, an efficiency-aware benchmark for dental multimodal question answering that brings together three datasets spanning approximately 1,159 patients, five task types and seven metrics. Across typical 14 VLMs, our results reveals an interesting observation: compact VLMs (e.g., 2B-parameter models) outperform larger VLMs in accuracy while requiring substantially lower computational costs in dental image understanding. Deployed locally on an iPhone 17 Pro, our finetuned compact VLM Pocket-Dentist-2B processed each sample in 4.31 s, reducing latency by 4.9-fold and memory use by 2.3-fold compared with a 7B baseline.

Problem

• dental image understanding
• multimodal large language models
• on-device deployment
• computational efficiency
• clinical prescreening

Innovation

• efficient multimodal LLM
• on-device inference
• dental image understanding
• compact VLM
• privacy-preserving screening