A systematic survey on image description techniques for STEM domains

The proliferation of visual data in Science, Technology, Engineering, and Mathematics (STEM) presents accessibility barriers for blind and low-vision users. While Artificial Intelligence (AI) can generate alternative descriptions of STEM images, research remains fragmented and pratical impact is limited. This systematic survey examines 20 peer-reviewed studies on AI-based STEM visual description, focusing on accessibility and Human–Computer Interaction (HCI). Following PRISMA methodology and a ROBIS-based risk-of-bias assessment, the review analyzes (i) STEM visuals targeted, (ii) AI architectures employed, (iii) datasets and evaluation metrics, and (iv) interaction modalities for delivering descriptions. Findings show a shift from static alt-text toward interactive, multimodal systems integrating conversational interfaces, keyboard navigation, audio, and haptic feedback. However, challenges persist, including hallucinations, limited accessibility-first datasets co-designed with BLV users, and overreliance on automatic text-overlap metrics. The survey identifies future HCI priorities: user-controlled verbosity, explainable AI pipelines, and integration of accessible description into mainstream STEM environments.