Recognition of Handwritten Medical Prescription using CNN Bi-LSTM With Lexicon Search
Aayush Razdan, Aditya Raghavan, Megh Panandikar, Arya Pol, Karina Hedaoo, Rajesh A. Patil · 2023
Digital technologies have advanced more rapidly than any innovation in the history of the world. The field of Medicine, however, seems to be lagging behind in this regard. A large number of medical prescriptions around the world are still handwritten on paper. A handwriting detection model for medical prescriptions could digitalize written prescriptions for users in a matter of seconds from a simple picture. Moreover, the problems of ambiguity and confusion in deciphering a doctor's handwriting would be solved. Handwritten Text Recognition (HTR) is implemented fundamentally using a combination of Convoluted Neural Networks (CNN) and Recurrent Neural Networks (RNN) on the popular IAM dataset of written words. This paper presents multiple improvements to the current approach by introducing real-life written prescription data as the input, as well as tuning hyperparameters for more accurate results. Lexicon Search has been implemented as a decoding algorithm that checks output predictions against a vast database of drug names used in India. This makes the model work better specifically for medical prescription data in comparison to basic models.