Assessment of infection levels in mechanical circulatory assist devi̇ce implanted patients utilizing image processing and deep learning methods

In this study, infection levels of possible driveline complications due to the implantation of mechanical circulatory assist devices were investigated with image processing and deep learning methods. Driveline infection due to the left ventricular assist devices (LVAD) implantation is the third most common cause of death in patients. Driveline infection is the manifestation of symptoms such as edema, warmth, purulent discharge in the skin tissue. These symptoms cause lesions in the skin tissue. In this study, image processing techniques and deep learning methods were used to detect driveline infection that may occur after implantation of LVAD due to the heart failure. Within the scope of the study, infection status was examined with images taken from patients who underwent mechanical circulatory support system implantation due to the heart failure. For these determinations, first of all, the exit region of the driveline from the body tissue was detected. In the detection of the body exit region of the driveline cable, segmentation was performed on the skin tissue. The K-means algorithm was chosen for the segmentation process. Region of the detected driveline exit-line was used as an input for the feature extraction and deep learning operations. Three different methods which are wavelet transform, Shannon entropy and gray level co-occurrence matrix were utilized in the feature extraction stage. The inferences have been made depending on these methods. A deep learning method was used in addition to the feature extraction method. Convolutional neural network architecture was developed and used for deep learning to predict the infection level of the patients. The success rate of the model, which was developed depending on the existing data structure and amount, was calculated as 90%.