Real-time edge tracking of deep blood vessels in ultrasound videos

A wide range of technologies have been developed in the last decade to simplify the work of the health department and a multitude of applications were born to improve diagnoses. This thesis aimed to develop a real-time deep blood vessels tracking system from an online echography video stream. Nowadays ultrasonography is even more important in the process of diagnosis of a great number of diseases. However, the lack of standardization represents a remarkable limitation to the diffusion of this technique, and the methodology is still user-dependent. A technology based on image processing algorithms has been developed and patented at Politecnico di Torino in 2018 to support the clinician in the measurement of inferior vena cava (IVC), trying to attenuate the problems mentioned before. The original patented algorithm was originally implemented in MATLAB and was too slow to be used in a real-time application (mean±std of 3±1.9 frames per second analyzed on a pool of 10 different videos), on a calculator with Intel Core i7-10750H with base frequency 2.6 GHz, 16GB RAM, and Microsoft Windows 10. The slowness was caused by sections of code with exponential complexity. The main constraint in a real-time application is the input-output latency that must be short as possible. To achieve this goal, the code was optimized, and an intensive use of multithreading programming in critical sections helped in improving performance. The maximum complexity of the code in the critical section is now quadratic, and the programming language is written in C++. The number of frames per second analyzed and printed was 33 (±std=2.6 on a pool of 10 different videos) on the same calculator mentioned before. The RAM usage is low, with an occupation up to 4 GB per hour of video. Videos in the analysis usually last 60 seconds, thus they can run on portable devices like tablets or low-end laptops without significant limitations. The application works on three-layers: a Front-End, a Back-End, and a low-level library that interacts with third parts hardware. This organization in layers helps to manage software (maintenance or reading) and improve it with less complexity independently from the hardware and the user interface (UI). Other design choices are explained in the dissertation with a detailed analysis of different possible scenarios, and further implementations. The UI has been designed to simplify the interaction with a complex instrument. The user must follow a well-defined pattern and only a limited amount of parameters can be modified: this apparent limitation was introduced to help the doctors, who during the exam have one hand on the probe and the other one on the computer and could be forced to take the measure in emergency conditions. It is also a touch-friendly interface since it supports Microsoft Surface or another Windows-based touch solution. The UI was designed on Windows Form: this is a native approach on Windows, and it helps the user because it is OS-like. It is compatible only with Windows OS, which is widely adopted by Italian hospitals. This project can be used immediately to improve ultrasound acquisition quality giving the user feedback on the vessel position. This approach can also improve the speed of the acquisition reducing the waste of time for doctors. The extracted data is the starting point for a standardization process instead of the actual qualitative and subjective measurement, and it will enhance future diagnosis outcomes.