MedGemma Fine-Tuned for Medical Equipment Maintenance QA in Low-Resource Settings

A recent study published on arXiv (2608.08896) presents a framework for question-answering regarding the maintenance of multi-modality medical equipment, enhancing the MedGemma-4b-it model to aid in technical troubleshooting within low-resource environments. This research, informed by a multi-country survey involving nine low- and middle-income countries (LMICs), compiled technical manuals from MRI and ultrasound devices to develop the INGENZI_DatasetV1, which includes 10,294 carefully filtered QA-context pairs. By employing QLoRA-based parameter-efficient fine-tuning, the model was optimized to analyze system error logs and produce detailed repair instructions. The fine-tuned model demonstrated significant advancements over the baseline, with the F1 score rising from 0.22 to 0.38 and ROUGE-2 from 0.18 to 0.41. This work addresses the pressing issue of imaging device downtime in LMICs, often caused by a lack of specialized biomedical engineering support.