Blood loss can occur in any surgical procedure. However, certain types of surgery, such as open orthopedic surgery, are associated with blood loss of a magnitude that may lead to the need for transfusion. Hereby, the patient’s blood loss should be minimized to reduce the patient risk, need for blood transfusion and hospitation time and improve the outcome of the intervention. In our research project, we are aiming to quantify the patient’s level of bleeding and accumulated blood loss volume in realtime and use a sonification approach to create awareness by continuously informing the operating surgeon about the current bleeding situation through acoustic signals. Therefore, we integrated an Arduino-based wage sensor into the surgical suction device. However, in an orthopedic surgery flushing solution is used to clean the wound during intervention. To quantify the patient’s blood loss in the suction system it is therefore crucial to automatically identify the flushing and the resulting dilution of the blood.
This Master Thesis should extend our current measurement setup with a computer vision based system for the liquid stream classification in the transparent suction tube. The first task will be characterized by designing the camera setup and choosing the hardware for data acquisition. The camera will be mounted in a fixed position to the tube and will observe the liquid flow. A deep learning based system should be developed and trained to classify the liquid stream into different classes for intensity and blood concentration. This work includes the CAD design of a camera mount and further structure components which will be manufactured with our in-house 3D printers. Finally, the solution should be integrated into our Python-based host application.
Materials/Methods/Tools:
Deep Learning Framework of choice (Tensorflow/Pytorch etc.), MATLAB / Python, CAD design and 3D printing
Blood loss can occur in any surgical procedure. However, certain types of surgery, such as open orthopedic surgery, are associated with blood loss of a magnitude that may lead to the need for transfusion. Hereby, the patient’s blood loss should be minimized to reduce the patient risk, need for blood transfusion and hospitation time and improve the outcome of the intervention. In our research project, we are aiming to quantify the patient’s level of bleeding and accumulated blood loss volume in realtime and use a sonification approach to create awareness by continuously informing the operating surgeon about the current bleeding situation through acoustic signals. Therefore, we integrated an Arduino-based wage sensor into the surgical suction device. However, in an orthopedic surgery flushing solution is used to clean the wound during intervention. To quantify the patient’s blood loss in the suction system it is therefore crucial to automatically identify the flushing and the resulting dilution of the blood.
This Master Thesis should extend our current measurement setup with a computer vision based system for the liquid stream classification in the transparent suction tube. The first task will be characterized by designing the camera setup and choosing the hardware for data acquisition. The camera will be mounted in a fixed position to the tube and will observe the liquid flow. A deep learning based system should be developed and trained to classify the liquid stream into different classes for intensity and blood concentration. This work includes the CAD design of a camera mount and further structure components which will be manufactured with our in-house 3D printers. Finally, the solution should be integrated into our Python-based host application.
Materials/Methods/Tools: Deep Learning Framework of choice (Tensorflow/Pytorch etc.), MATLAB / Python, CAD design and 3D printing
- Design of the camera setup
- Definition of hardware for data acquisition
- Developement of deep learning based system for liquid stream classification
- Design of the camera setup - Definition of hardware for data acquisition - Developement of deep learning based system for liquid stream classification