A Deep Learning Ai-based System For Drug Identification

• By: MEI, Nai-Hwa (China Taiwan)
• Co-author(s): Ms Nai-Hwa Mei (National Yang Ming Chiao Tung University, Taipei, China Taiwan)
  Mr Chen-Yao Ho (Shin Kong Wu Ho-Su Memorial Hospital, Taipei, China Taiwan)
  
• Abstract:

  Background:
  Medication errors harm at least 1.5 million people every year leading to an estimated $3.5 billion in morbidity and mortality costs annually. Whether it be the increased number of medications, older age, or poor transition of care that is associated with medication errors, we have continued to search for better methods of reducing said harm. Hence, the development of a deep learning-based system to accurately identify prescription pills.
  
  Purpose:
  To propose the utilization of artificial intelligence (AI) in identifying numerous pills with high precision. In hopes to reduce patients’ misuse of medications and assist healthcare professionals.
  
  Method:
  165 pill images were collected from Shin Kong Wu Ho-Su Memorial Hospital’s database for identification. The confusion matrix concept within the ResNet50 model developed by Natalia Larios Delgado and EfficientNet-b5 training model was then adopted for image classification and text detection. This study then trained and compared the proposed models based on images of the front, back, and side views of drugs.
  
  Results:
  The experimental results show that with 165 images generated from 11 classified as “Grade 3 confusion matrix” drugs the AI-based system achieved an accuracy level of 90.9%.
  
  Conclusion:
  This study demonstrated how a deep learning AI-based model for pill identification can generate accuracies greater than 90%. If integrated into existing prescription systems such as hospitals, community pharmacies, or other healthcare facilities, not only does it reduce medical errors but also allows pharmacists time to conduct higher-level tasks by simplifying the drug identification process.