Chromosome Cluster Type Identification Using a Swin Transformer

The analysis of chromosome karyotypes is crucial for diagnosing genetic disorders such as Patau syndrome, Edward and Down syndrome. Chromosome cluster type identification a key step in the automated karyotypes. State-of-the-art cluster-type techniques are based on convolutional neural networks (CNNs) fail to exploit global context. To address this limitation state art, paper proposes transformer network, (CCT), that exploits swin backbone successfully captures long-range dependencies image. Additionally, we find proposed CCT has large number model parameters, which makes it prone overfitting (small) dataset images. alleviate limited availability training data, also utilizes transfer learning approach. Experiments demonstrate outperforms state-of-the-art methods well traditional vision transformer. Furthermore, provide insights improved performance, activation maps obtained using Gradient Attention Rollout.