Status : Verified
| Personal Name | Ambita, Ara Abigail E. |
|---|---|
| Resource Title | Machine learning estimation of intra-operative blood loss from gauze using image features from visible and near-infrared wavelengths |
| Date Issued | June 2022 |
| Abstract | Blood loss estimation during surgery is essential in determining the appropriate transfusion decisions. Triton System estimates intraoperative haemoglobin (Hb) loss using AI but is unable to directly assess the amount of blood loss. While some studies have developed a method for direct blood loss assessment, non-sanguinous fluids that largely contribute to incorrect assessment is not considered. Due to blood’s IR light absorption, Near Infrared (NIR) spectroscopy is widely applied to blood-related analysis. Hence, to achieve a cheap, direct, and accurate blood loss estimation, a technique that combines conventional imaging and spectroscopy, is evaluated. Specifically, we aim to investigate the effects of dual wavelength - visible (VIS) and near-infrared (NIR) on blood loss estimation. The technique is applied on a dataset containing pure and diluted blood as well as wet or dry gauze samples. Precise estimations were achieved using the the dual Vis-NIR wavelength, rendering a +28.30%, +48%, +27.97%, and 25.72% improvement on the MAE, MSE, RMSE, and MAPE, compared to using a single Vis wavelength. The model further improved when fusing spectral and spatial features extracted from multiple wavelengths (Vis, NIR, dual Vis-NIR) but with 49.50%, 72.80%, 48.10%, and 49.77% improvement on thesame metrics. Acceptable mean absolute percentage error (MAPE) of 8.475% is produced which is within the range of maximum allowable error of 20%. In conclusion, this study demonstrates that volumes < 1g and diluted blood > 90% water content are detected in IR that consequently improves the estimation. |
| Degree Course | MS Computer Science |
| Language | English |
| Keyword | machine learning; blood loss estimation; computer vision; AI for healthcare |
| Material Type | Thesis/Dissertation |
Preliminary Pages
185.04 Kb
Category : I - Has patentable or registrable invention of creation.
Access Permission : Limited Access