Comet Former: A Novel Deep Learning Architecture using the U-MixFormer Model to optimize Comet Assay Segmentation

Increased levels of DNA damage and ineffective repair mechanisms are the underlying bio-molecular events in the pathogenesis of most life-threatening diseases like cancer and degenerative diseases. The comet assay is a widely used method to quantify the amount of DNA damage from chemotherapeutic and radiotherapy agents. However, current methods for quantifying DNA damage, which involve manual segmentation and identification of comet areas, are time-consuming and inefficient. New Deep Learning models are used to complete this process as they are much more effective. To address this limitation, we propose a novel deep learning-based architecture, CometAI, which utilizes the U-MixFormer model to segment and identify comets in Comet Assay images efficiently. CometAI achieved high scores in intersection over union(IoU) and F1 Score, essential image segmentation metrics. Compared with current segmentation tools, CometAI offers a significant improvement in segmenting both the comet head and tail, allowing for more accurate DNA damage assessment in various fields.