Synovial fluid imbalance in joints plays a significant role in the diagnosis of Rheumatoid arthritis (RA) at an early stage. RA mostly attacks the small joints like finger and wrist joints which makes it challenging to segment the synovial fluid from those small joints automatically. Although ultrasonography (USG) imaging is very sensitive to small joints and its fluid assessment, segmentation of synovial fluid regions from the USG images in the literature are less understood. Moreover, towards computer vision related research (especially segmentation of the suspicious abnormal regions) using USG imaging, several challenges exists including (a) USG images are prone to the certain artifacts in terms of noises because of which the presence of the different appearance of synovial fluid is less distinguishable with respect to the other anatomical appearances, (b) Also, with respect to imbalance occurrence, synovial fluid changes its shape, size and locations from one USG image to another USG image. To cope with such pre-defined challenges, we proposed a novel lightweight network named as “Synovial Fluid Region Segmentation Network (SFRSeg-Net)” for segmentation of synovial fluid regions from the USG images. The proposed network enriches by incorporating a novel Interpretable Element Wise Additive-Discrete Wavelet Transform (IEWA-DWT) based down sampling strategy to extract the significant salient features by ignoring noise imposed in USG imaging and maintain the original image integrity. As synovial fluid varies its shape in one image to another image, so boundary loss is also significant learning parameter with pixel wise region loss and its mutual combination is used as a loss function in our proposed network. Experimental results on publicly available USG imaging dataset reveal that our proposed SFRSeg-Net performed well with Dice similarity coefficient of 0.9066 ± 0.0520 which surpasses both the most recent state-of-the-art techniques and the current baseline.