Blood Loss Estimation on Surgical Sponges via Machine Learning and Rate-Based Pixel Counting Method

Accurate measurement of intraoperative blood loss is crucial for the success of surgical procedures and ensuring patient safety. However, current methods, such as visual estimation and gravimetric techniques, have limitations in terms of accuracy and practicality. Visual estimation, in particular, requires extensive experience typically possessed by expert consultants. To address these limitations, this paper introduces a machine learning (ML) and image processing-based approach to predict blood loss on surgical sponges without the need for expensive equipment or specialized expertise. The system analyzes sponge images captured using a standard mobile phone camera. We propose a novel approach that generates a histogram to estimate the proportion of red-toned pixels within the stain, relative to the total number of pixels in the sponge, with this rate-based pixel count offering a more accurate color-based assessment of blood volume. We trained our model using a dataset of sponge images with varying blood levels from Aydın Adnan Menderes University Hospital. Our experiments demonstrate that the method successfully predicts different blood volumes, achieving a minimum RMSE of 0.91 and a maximum R2 of 0.93 with Gradient Boosting, while Artificial Neural Networks yielded similar results.