Two-stage hemoglobin prediction based on prior causality.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Perioperative hemoglobin (Hb) levels are critical for tissue metabolism and guide intraoperative blood transfusions.
• Reduced Hb levels (anemia) during surgery can lead to severe complications and mortality.
• Accurate, non-invasive Hb prediction is crucial for patient safety.

Purpose of the Study:

• To develop a non-invasive model for predicting perioperative hemoglobin concentration.
• To leverage deep learning and a priori causal knowledge for precise Hb level estimation.
• To create an automated framework for Hb prediction using palpebral conjunctiva images.

Main Methods:

• Deep neural semantic segmentation and convolutional neural networks were employed.
• A priori causal knowledge guided the positioning of the palpebral region.
• A regression prediction model was built using a neural network on the identified region.

Main Results:

• The developed model achieved an R-squared value of 0.512.
• The explained variance score reached 0.535.
• The mean absolute error for Hb concentration prediction was 1.521.

Conclusions:

• A deep learning model was successfully constructed to predict eyelid Hb in perioperative patients.
• The model utilizes a priori causal knowledge for accurate Hb concentration prediction.
• This non-invasive approach holds significant potential for clinical applications in managing perioperative anemia.