Auto-completion of medical terms using character level NLP.
Overview
As a minor technical project at my undergraduate university under the guidance of Prof. Naveen Aggarwal, we decided to work on Auto-completion of medical terms using character level NLP. We chose this project because it exposed us to two crucial machine learning applications, i.e. image processing and Natural language processing.
Context
The Covid-19 pandemic has brought to the fore the critical need for cutting-edge technological tools and innovation in the areas of public health, medicine and wellness. One such technology is telemedicine and digital prescriptions, which allows patients to receive follow-ups at home, thus avoiding the spread of COVID-19 in overcrowded emergency or waiting rooms. As paper prescriptions have been readily replaced by digital one's, it has created a need for cost-effective and efficient electronic systems to understand handwritten words and even provide suggestions to complete them.
Advantages
A prescription written by the doctor is a critical component of any clinical check-up. Thus, if saved, the time spent writing these lengthy prescriptions can be used to attend to other cases, thereby increasing the doctor's efficiency.
Methodology
We aimed to develop an AI tool to make the process of providing the digital prescription be a hassle-free task for doctors. For this purpose, we developed an integrated algorithm that includes:
Recognising handwriting employing a pre-trained ResNet-50 Architecture with imagenet weights.
Character level NLP model to complete the prescribed drug.
Web Application made using Python and Flask to showcase the results.
Result
The final paradigm requires the input of a few letters to predict the complete word, thus saving the doctor's time and effort and increasing the efficiency. For the purpose of demonstration of the predictive algorithm, we used the handwritten image as an input.
