Deep learning analysis of particle content in extracted slow-release morphine: longer boiling reduces large fragments while retaining morphine extraction

INTRODUCTION: Injecting drug users often extract morphine from slow-release oral tablets, potentially leading to harmful particle contamination upon injection. This study assesses the efficiency of morphine extraction and particle content of filtrates produced by various methods employed by drug users in Trondheim, Norway. The findings provide important insights that can inform harm-reduction services and healthcare providers in efforts to reduce injection-related morbidity among people who already inject drugs. METHODS: Four extraction methods were evaluated using 60 mg Dolcontin tablets: Method A (no coating removal, 3-minute boiling), Method B (coating removal, crushing, 3-minute boiling), Method C (coating removal, 3-minute boiling), and Method D (coating removal, 10-minute boiling). Resulting solutions were filtered using cotton balls, and morphine content was quantified using LC-MS/MS. Particle content of filtrates was analyzed using slide scanning, deep learning-based particle segmentation, and QuPath image analysis software. RESULTS: Morphine recovery ranged from 81.2% (Method D) to 91.3% (Method B). Method A yielded a significant presence of small insoluble particles (500 μm). Method C exhibited the highest density of medium-sized particles (100-500 μm). Method D generated the fewest particles across all size categories. CONCLUSION: The extraction methods used by injecting drug users result in significant variability in morphine recovery and particle content of filtrates. Method D (10-minute vs. 3-minute boiling) demonstrated the highest efficiency in particle reduction, with only 10% less morphine recovery. Lack of coat removal significantly increases the number of primarily small (