FILET VEINEUX PROFONDE : IDENTIFICATION DE LA THROMBOSE VEINE PROFONDE VIA UN RÉSEAU D'APPRENTISSAGE PROFONDE OPTIMISÉ PAR LA STERNE fuligineuse

Abstract

Deep vein thrombosis (DVT) occurs when thrombosis (blood clots) forms in veins far below the skin's surface due to veins or sluggish blood flow injuries. An obstruction in blood flow through a vein could be partially or completely caused by blood clots. DVTs typically occur in the thigh, lower leg, or pelvis, yet can also occur in other body parts, such as the brain, liver, intestines, arm, or kidney. This research proposes a novel Deep Vein Net, integrating deep learning-based dilated CNN and Sooty Tern optimization to detect DVT from CT and MRI images efficiently. The input CT and MRI images are pre-processed to eliminate noise artifacts using the Discrete Wavelet Transform (DWT). Furthermore, the pre-processed images are fed into a Dilated convolutional neural network (Dilated CNN) for feature extraction to extract the most pertinent features. Lastly, the STO algorithm uses the fuzzy Extreme Learning Machine of thrombosis stages normal and DVT to select the best features for additional classification. Metrics like rec, spe, acc, pre, and F1 scores were used to assess the Deep Vein Net's performance. The suggested method achieves a classification accuracy of 99.25 % when identifying DVT cases.