Segmentation of medical images is a vital component in clinical operations, facilitating precise diagnostic procedures, development of treatment methodologies, and monitoring of disease progression. However, challenges arise from diverse medical image quality and resolution, influenced by factors like noise, artifacts, and imaging protocols. Despite advancements in deep neural networks, accurately delineating complex anatomical structures with subtle borders and overlapping tissues remains a challenge, particularly in achieving precise boundary segmentation. To overcome this constraint, we introduce a novel ‘3D Inverseform Loss’ function to capture the spatial distance between the prediction and the ground truth boundary maps by integrating inverse transformation network, thereby enhancing the delineation of predicted segmented boundaries. This approach significantly improves the precision of segmentation, especially in complex and non-rigid anatomical structures. Extensive experiments and comparative study between the current SOTA model and the proposed model indicate that the proposed model outperforms the existing SOTA model by \(2.05\%\) , particularly in 3D abdominal segmentation and achieved a dice coefficient of \(91.02\%\) .