Detecting the required portion from the medical images requires complex handling. This proposed work suggests the deep learning methodology to detect the knee structure from the medical image. Deep learning techniques such as U-Net, U-Net++, PSPNet, and AUNet was used to detect the required region of interest from the input medical images. Starting with image edge detection, segmentation, and object identification based on edges, U-Net offers a high efficiency when compared with the similar models. This study offers an efficient and accurate results in terms of memory, processing speed, and number of operations per second. Each model was tested individually with the same dataset under various constraints. With the results obtained, PSPNet achieved better results with a better outcome of over 85%. This suggests that PSPNet is suitable for segmenting medical images based on the region of interest. U-Net++ and AUNet follow closely behind PSPNet this showing that PSPNet offers better performance. Also, PSPNet is the preferred model for accuracy-influencing scenarios. The findings obtained in this study show that PSPNet met the highest segmentation accuracy among the evaluated models, achieving a Mean Intersection over Union (mIoU) of 85.4%, which highlights its strong capability in capturing multi-scale contextual features and accurately identifying the region of interest in knee medical images.