Leveraging Deep Learning for Automated Image Segmentation and Accurate Quantification of Western Blot Protein Bands: A U-Net Approach for Semantic Segmentation

Background and Aims: Traditional Western blotting has provided qualitative information, whereas automation in Western blot image analysis represents a significant step toward efficient and effective protein band extraction. Several conventional approaches are usually manually processed; hence, they are time-consuming and prone to human bias. This study develops an automatic Western blot imaging method that uses U-Net, performing semantic segmentation for quantitatively identifying protein bands. Methods: The proposed approach applies semantic segmentation to Western blot images using the deep learning architecture U-Net. ۶۸ Western blot images were used to evaluate the effectiveness of the automated segmentation method. Results were obtained using a total of ۲۱۴ images after augmentation, which included ۶۸ original images. Performance assessments have been conducted concerning multiple metrics, including accuracy, precision, sensitivity, F۱-score, specificity, Dice index, and Jaccard index. Results: The proposed method provided an accuracy of ۹۰.۷%, precision of ۸۲.۷%, sensitivity of ۸۶.۱%, F۱-score of ۸۳.۷%, specificity of ۸۶.۴%, Dice index of ۸۳.۷%, and Jaccard index of ۷۷.۲% for the augmented dataset with ۲۱۴ images, while in the original dataset, having ۶۸ images, it provides improved results of accuracy by ۹۴.۰%, precision of ۹۳.۵%, sensitivity of ۸۹.۷%, F۱-score of ۹۰.۵%, specificity of ۸۹.۷%, Dice index of ۹۰%, and Jaccard index of ۸۴.۵%. Conclusion: The automated U-Net-based method performed very well in segmenting Western blot images, both with augmented and original data, with high accuracy, precision, and similarity metrics. This method shows promise for improving efficiency and reproducibility in protein band detection related to Western blots.

کلیدواژه ها:

Western Blotting ، Image Processing ، Automated Segmentation