A Novel Handwritten Prescription Recognition with Stochastic Gradient Descent Using Adaptive Momentum Learning

Personalized medicine relies heavily on handwritten prescriptions since they provide patients with instructions that are customized to them. In the medical industry, accurately interpreting handwritten prescriptions poses a serious difficulty that affects patient safety and the standard of treatment due to the illegible handwriting by doctors. This research presents “MediScriptNet”, a novel model that improves handwritten medical prescription detection by utilizing the advantages of both Bidirectional Long Short-Term Memory networks (BiLSTM) and Depthwise Separable Convolutional Networks (DSConvoNet). MediScriptNet establishes a new benchmark for handwritten prescription decoding complex cursive scripts by merging the sequence prediction expertise of BiLSTM with the lightweight and effective feature extraction skills of DSConvoNet. We introduce a novel architecture with an adaptable momentum learning rate that is optimized by stochastic gradient descent (SGD) and Particle Swarm Optimization (PSO). More stable and quicker convergence can be achieved by using the adaptive momentum to enable subtle adjustments to the learning rate based on the complexity of the recognition patterns and the model’s historical learning curve. Validation on a wide range of prescription datasets shows that MediScriptNet performs much better than current models in terms of accuracy, speed, and computing efficiency with performance of 94.5% accuracy as adaptive learning enhances the SGD optimization.