aspg-2-2967-1791419336 20261008002856 American Scientific Publishing Group admin@americaspg.com American Scientific Publishing Group Journal of Cybersecurity and Information Management JCIM 2769-7851 2690-6775 2024 14 2 Koppagiri Jyothsna Devi School of Computer Science and Engineering, VIT-AP University, Andhra Pradesh, India Gouranga Mandal School of Computer Science and Engineering, VIT-AP University, Andhra Pradesh, India For biomedical image analysis, instance segmentation is crucial. It is still difficult because of the intricate backdrop elements, the significant variation in object appearances, the large number of overlapping items, and the hazy object borders. Deep learning-based techniques, which may be separated into proposal-free and proposal-based approaches, have been frequently employed recently to overcome these challenges. The existing approaches experience information loss due to their concentration on either local-level instance features or global-level semantics. To solve this problem, this work proposes an improved dense Net ( style='mso-bidi-font-style:normal'> "Cambria Math",serif;mso-fareast-font-family:Calibri;mso-fareast-theme-font: minor-latin;mso-bidi-font-family:"Times New Roman";mso-ansi-language:EN-US; mso-fareast-language:EN-US'>ID 10.0pt;font-family:"Cambria Math",serif;mso-fareast-font-family:Calibri; mso-fareast-theme-font:minor-latin;mso-bidi-font-family:"Times New Roman"; mso-ansi-language:EN-US;mso-fareast-language:EN-US'> m:val="p"/>-Net ) that mixes instance and semantic data. The suggested normal'> mso-fareast-font-family:Calibri;mso-fareast-theme-font:minor-latin;mso-bidi-font-family: "Times New Roman";mso-ansi-language:EN-US;mso-fareast-language:EN-US'>ID style='font-size:10.0pt;font-family:"Cambria Math",serif;mso-fareast-font-family: Calibri;mso-fareast-theme-font:minor-latin;mso-bidi-font-family:"Times New Roman"; mso-ansi-language:EN-US;mso-fareast-language:EN-US'> m:val="p"/>-Net promotes the acquisition of semantic contextual information by the instance branch by linking instance prediction and semantic features via a residual attention feature integration strategy. The confidence score of each item is then matched with the accuracy of the prediction using a dense quality sub-branch that is created. A consistency regularisation technique is also proposed for the robust learning of segmentation for instance branches and the semantic segments tasks. By proving its utility, the proposed style='mso-bidi-font-style:normal'> "Cambria Math",serif;mso-fareast-font-family:Calibri;mso-fareast-theme-font: minor-latin;mso-bidi-font-family:"Times New Roman";mso-ansi-language:EN-US; mso-fareast-language:EN-US'>ID 10.0pt;font-family:"Cambria Math",serif;mso-fareast-font-family:Calibri; mso-fareast-theme-font:minor-latin;mso-bidi-font-family:"Times New Roman"; mso-ansi-language:EN-US;mso-fareast-language:EN-US'> m:val="p"/>-Net outperforms prevailing approaches on various biomedical datasets. 2024 101 114 2967 https://creativecommons.org/licenses/by/4.0/ 10.54216/JCIM.140207 https://www.americaspg.com/journal/2/article/2967 https://www.americaspg.com/storage/11720055156.pdf