Accurate localisation plays a crucial role in automated diagnostics within medical imaging analysis. Colorectal disease diagnosis and treatment plan may be impacted by accurate localisation of the sigmoid colon. This work explores different Convolutional Neural Networks (CNN) architectures to improve the accuracy of sigmoid colon localisation in 3D computed tomography (CT) image data. Our study explored a range of CNN configurations, including variances in layers and parameters, to analytically evaluate their effect on localisation accuracy. From our comparative analysis, we identified key architectural features that improve localisation accuracy. Based on these findings, we developed a novel 3D CNN architecture, named Sigloc 3D CNN, that is specifically adapted for efficient sigmoid colon localisation. Our model includes optimised layer configurations and parameters, leading to a significant improvement in localisation accuracy. This work used sigmoid colon 3D CT image data collected from patients undergoing treatment for diverticulitis. For evaluation, we used Intersection over Union (IoU) and Mean Absolute Error (MAE) as the primary metrics, and MAE as the loss function in the CNNs. Based on benchmarked architectures, our Sigloc 3D CNN model exhibited superior performance, achieving the highest IoU of 0.6946 ± 0.1251 and the lowest MAE 0.0121 ± 0.0039. These outcomes underscore the effectiveness of the proposed Sigloc 3D CNN model in accurately localising the sigmoid colon, providing a valuable tool for advanced medical image analysis, and contributing to the broader field of computer-aided diagnosis systems.