AgsNet: An Attention-Guided Lightweight Segmentation Network

Urinalysis test strips are commonly used for urine routine examination. However, due to possible defects in the liquid path, such as blockages, droplets may leak during the process of dropping urine samples onto the test strips, which severely affects the results of medical tests. Therefore, we propose the Attention-guided segmentation Network (AgsNet) to address errors in medical test results caused by defects in the liquid path. AgsNet adapts its focus to different areas of the test strip image, effectively extracting a richer and more diverse set of features. The best segmentation result is obtained with the AgsNet achieving a mean Intersection over Union (mIoU) score of 71.8 and mean Average Precision (mAP) scores of 84.49, respectively. These results underscore AgsNet’s potential in significantly reducing the impact of liquid pathway defects on the reliability of urinalysis test outcomes.