Deep learning has revolutionized various domains, showing great performance for several perceptual tasks, in the fields of computer vision, speech recognition, and natural language processing. However, designing optimal deep neural network (DNN) architectures often rely on expert knowledge and time-consuming trial and error approaches. Neural Architecture Search (NAS) has emerged as a promising solution, automating the design process to discover architectures that enhance performance and efficiency. This work introduces SegQSNAS, an extension of SegQNAS, designed for Q-NAS (Quantum-inspired Neural Architecture Search) in semantic segmentation, particularly within the medical imaging domain. SegQSNAS is designed to search for symmetrical U-Net-like architectures, thereby reducing the search space and eliminating the need for feasibility checks that were needed in SegQNAS. Furthermore, it enhances the search process by incorporating a two-point crossover operation in order to improve the exploitation during evolutionary process, alongside the addition of self-attention, MobileNet, and EfficientNet functions to the search space that enables the discovery of efficient, customized networks. To address inaccuracies in multi-class scenarios, the SegQNAS implementation of Dice-Sorensen Coefficient (DSC) and loss functions were corrected because it provides an overestimated DSC score in those scenarios. Experiments on medical segmentation datasets from the Medical Segmentation Decathlon challenge demonstrate that on the prostate dataset experiment, SegQSNAS performed well with the best results in DSC score, parameter count and GPU days, archiving 0.7924 in average DSC score with half million parameters. It also archive of good results on the liver dataset experiment due to the best trade-off considering that it is a limited computing resource scenario. Although, in some experiments it show some limitation of the search strategy or computational resource available. In addition, in any experiment the self-attention block was not selected in the best architecture found, and this is indicative that the predefined maximum number of nodes or the complexity of the tested problems might not have been high enough for this operation to be selected during evolution process.