Deep-learning-based cross-modality conversion of bright-field images for nucleus segmentation

Achieving nucleus segmentation based on bright-field microscopic images is a fundamental technique for quantitative biological analysis. Traditional segmentation methods exhibit limited performance due to challenges such as low contrast in bright-field images. This study proposes a deep learning-based cross-modality conversion technique for nucleus segmentation using bright-field images. We employ U-Net to establish a cross-modality mapping paradigm for bright-field-to-fluorescent image conversion. Subsequently, the channel attention mechanism in U-NetV2 is integrated to focus on the critical nuclear features. Experimental results demonstrate that the proposed method achieves a Dice coefficient of 0.90±0.03 on the test set, outperforming U-Net and PraNet by 2% and 13%, respectively, while reducing the Average Symmetric Surface Distance (ASD) to 2.06. This approach combines the convenience of bright-field imaging with the high quality of fluorescent images, providing a robust solution for dynamic live-cell monitoring.