Mitochondrial Segmentation in Microscopy Images Using UNet-VGG19

Mitochondria are essential organelles crucial for energy generation and cellular functions. Accurate segmentation of mitochondria from microscopy images is vital for understanding their morphology and functionality. This paper presents an enhanced segmentation method that integrates U-Net and VGG19 architectures, leveraging their combined strengths to improve precision. Previous studies in mitochondrial segmentation faced challenges such as low accuracy in boundary detection, insufficient diversity in training data, and less suitable loss functions. Traditional methods often failed to accurately distinguish complex mitochondrial structures. The proposed approach in this paper combines U-Net and VGG19 architectures and employs diverse data augmentation techniques, resulting in improved model generalization and accuracy. The use of the Jaccard loss function also enhances the training process. Tested on real datasets, the proposed method achieved an impressive IoU of 91% and a loss value of 0.02%, significantly outperforming existing methods and setting a new benchmark in mitochondrial segmentation.