MyoRCB-CAM-Seg: A New Supervised Segmentation Mechanism for Diagnosis of Myositis medical ultrasound images Integrating new Modified Convolutional Residual Blocks and Convolutional Attention Mechanism

Abstract

Myositis is an extremely uncommon condition that causes muscle weakening due to an immune system attack on muscle cells. An accurate and timely diagnosis is essential for successful treatment. Differentiating among various kinds of myositis remains a big problem, leading to incorrect diagnoses and delayed treatment. Segmentation can improve the visualisation of medical ultrasound images by highlighting the region of interest and reducing the image's complexity by reducing the quantity of data that must be processed. This research presents a a unique supervised segmentation architecture that can efficiently learn from a fair amount of medical images to efficiently perform accurate segmentation tasks at low computing cost. Our model utilises a novel encoder-decoder architecture that incorporates an attention mechanism and GroupNormalizaton into a residual convolution block, allowing it to acquire and analyse image data at several resolutions within the encoder segment. On a benchmark set of myositis ultrasound pictures, the suggested method achieves state-of-the-art performance across all metrics including Precision, Recall, mean Dice coefficient, Jaccard index, and Accuracy. The proposed method exhibits high generalisation capabilities, producing outstanding results despite having access to a modest sample of training data. The proposed model achieved remarkable improvement in terms of accuracy compared to Unet++, DeepLabV3 and the original duck architecture which reached up to 12 %, 7 % and 2 % respectively. In addition, the proposed model also accomplished a noticeable improvement in terms of Dice Coefficient extended to 13 %, 11 % and 5 % respectively, when compared to the traditional models Unet++, DeepLab V3 and the original duck architecture respectively.