We proposed a more advanced model for medical image segmentation in this article named as Adv-Unet++. Our design is essentially a deeply supervised based decoder–encoder architecture along with a nested skip connection in the dense network joined with the both decoder and encoder part of the model. The proposed skip connection is utilized to reduce the semantic discrepancy between the decoder and encoder sub-models’ generated feature-generated maps. We contend with the feature mappings from the decoder and encoder networks are semantically comparable, the optimization method would have a simpler time learning the procedure. Many medical image segmentation tasks include nodule segmentation in nuclei segmentation in microscopy images, low-dose CT scans of the chest, polyp segmentation in colonoscopy videos, and liver segmentation in CT scans of the abdomen. We have compared UNet and UNet++ with wider UNet techniques. Our tests show that an average IU gain of 3.4 and 3.9 points over broad UNet and UNet, respectively, may be achieved by UNet++ under deep supervision.