Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces fine-tuned small language models (Gemma-2-2B, Qwen3-1.7B, DeepSeek-R1-Distill-Qwen-1.5B) to detect the co-occurrence of body image distress, disordered eating, and metabolic challenges in social media posts related to PCOS. The models generate structured explanations with textual evidence, achieving 75.3% exact match accuracy in detecting these conditions. While effective for screening, performance declines with diagnostic complexity, limiting their use for autonomous diagnosis.
Open-source language models can now detect the hidden triple burden of body image distress, disordered eating, and metabolic challenges in PCOS patients from social media with surprisingly high accuracy.
Women with polycystic ovary syndrome (PCOS) face substantially elevated risks of body image distress, disordered eating, and metabolic challenges, yet existing natural language processing approaches for detecting these conditions lack transparency and cannot identify co-occurring presentations. We developed small, open-source language models to automatically detect this triple burden in social media posts with grounded explainability. We collected 1,000 PCOS-related posts from six subreddits, with two trained annotators labeling posts using guidelines operationalizing Lee et al. (2017) clinical framework. Three models (Gemma-2-2B, Qwen3-1.7B, DeepSeek-R1-Distill-Qwen-1.5B) were fine-tuned using Low-Rank Adaptation to generate structured explanations with textual evidence. The best model achieved 75.3 percent exact match accuracy on 150 held-out posts, with robust comorbidity detection and strong explainability. Performance declined with diagnostic complexity, indicating their best use is for screening rather than autonomous diagnosis.