Dual Model Medical Invoices Recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Medical Informatics

Background:

• Manual data entry of medical invoices is labor-intensive, with millions processed annually.
• Existing systems struggle with recognizing fragmented or breakpoint fonts common in invoices.
• Inefficient data entry leads to increased hospital costs and potential workflow disruptions.

Purpose of the Study:

• To develop an automated system for accurate medical invoice content recognition.
• To enhance the efficiency and reduce the cost of medical invoice processing in hospitals.
• To address the challenge of recognizing breakpoint fonts in invoice data.

Main Methods:

• A novel Gaussian blur and smoothing (GBS) preprocessing technique was developed to correct breakpoint fonts.
• An optimized Alexnet-Adam-CNN (AA-CNN) model was employed for initial character recognition.
• A recurrent neural network (RNN) was integrated as a semantic revision module to further improve accuracy.

Main Results:

• The GBS preprocessing effectively repaired breakpoint fonts in medical invoices.
• The optimized AA-CNN model demonstrated superior performance in recognizing breakpoint fonts compared to traditional CNNs.
• The combined GBS-CR method achieved an average increase of 10-15 percentage points in recognition rate compared to state-of-the-art methods.

Conclusions:

• The proposed GBS-CR method offers a significant advancement in automated medical invoice recognition.
• This system can substantially reduce manual data entry workload and associated costs for hospitals.
• The integration of specialized preprocessing and hybrid neural networks enhances recognition accuracy for challenging invoice data.