ChromEnhance-X: A Multi-Modal Deep Learning Framework for Chromosome Image Enhancement

Chromosome image analysis is crucial in cytogenetics, particularly for diagnosing genetic disorders and conducting cancer research. However, the low quality of these images often hampers accurate analysis. To address this, we introduce ChromEnhance-X, an advanced multi-modal deep learning framework that enhances chromosome image quality while preserving essential biological details. ChromEnhance-X integrates a conditional Generative Adversarial Network (cGAN) with a U-Net generator, a PatchGAN discriminator, a modified DnCNN for denoising, a self-attention module, and a learnable contrast enhancement network. This architecture tackles noise reduction, feature preservation, and contrast enhancement simultaneously. Our framework is designed with chromosome-specific considerations in mind, utilizing custom loss functions for band pattern preservation, karyotype consistency, and centromere detection accuracy. This ensures that the enhanced images retain critical biological features vital for precise cytogenetic analysis. We trained ChromEnhance-X on two datasets: the Cell Image Library (CIL) with 5,000 metaphase cell images, and the Bio Image Chromosome Classification Dataset (BICCD) with 42 chromosome objects. The results show significant improvements in image quality, with PSNR increasing by 25.3%, SSIM improving from 0.72 to 0.89, and FSIM rising from 0.81 to 0.93. Chromosome-specific metrics also improved markedly. Moreover, ChromEnhance-X is computationally efficient, processing images in under 50ms, making it ideal for real-time applications on resource-limited systems. Its robust performance across diverse datasets highlights its potential for broad applications in cytogenetics and biomedical imaging.