crossvit-covid19-fyp

COVID-19 chest X-ray classification uses convolutional neural networks to detect infection patterns. CrossViT, a dual-branch transformer architecture, processes X-ray images at multiple scales to capture both fine and coarse features, then outputs disease presence predictions. This Skill trains CrossViT-Tiny on the Rahman COVID dataset with CLAHE preprocessing and statistical validation including 95% confidence intervals and paired t-tests.

CrossViT training for X-ray classification requires significant GPU memory. This implementation runs on RTX 6000 Ada Generation hardware with 51GB VRAM. Memory constraints affect batch size, model depth, and training duration; insufficient VRAM will cause out-of-memory errors during transformer backbone computation.

X-ray preprocessing standardizes input quality and ensures reproducible results across runs. This Skill applies CLAHE (Contrast Limited Adaptive Histogram Equalization) to enhance contrast while preventing over-amplification of noise, preparing images for CrossViT input. Preprocessing is paired with rigorous statistical validation to meet reproducible-research standards.

Comparative evaluation validates model performance against known reference methods. This Skill evaluates CrossViT against five baseline models using McNemar and DeLong statistical tests, confidence intervals, and paired t-tests. Results quantify whether CrossViT outperforms alternatives on the Rahman COVID dataset with statistical significance.

Transformer X-ray classification requires deep learning libraries and image processing tools. This Skill depends on PyTorch for model training, OpenCV for image preprocessing, NumPy for numerical operations, and tqdm for progress monitoring. All dependencies are pinned to ensure reproducible environment setup across runs.

Reproducible-research practices ensure results are verifiable and replicable. This Skill provides dataset specifications, model architecture documentation, hardware constraints, preprocessing steps, baseline configurations, coding guidelines, and statistical validation procedures (confidence intervals, paired t-tests, McNemar, DeLong analyses) required for publication-quality COVID-19 classification research.