IEEE Xplore · ICIMA 2026

ConvNeXt with Coordinate Attention for chest X-ray diagnosis

Peer-reviewed computer vision work combining a ConvNeXt backbone with 1D Coordinate Attention for four-class pulmonary pathology classification.

DOI: 10.1109/ICIMA68728.2026.11564670 Doc ID: 11564670 Peer-Reviewed

Lung Disease Classification from Chest X-Rays using ConvNeXt with Coordinate Attention on the COVID-19 Radiography Dataset

Authors: Sasank Mangamuri, V. V. M. Rajavarapu, H. R. Channagiri, N. G. Thanubuddu, S. Rao P.

Accurate automated identification of pulmonary conditions from chest radiography requires fine-grained spatial features without heavy self-attention computational cost. We augment ConvNeXt with Coordinate Attention — applying 1D pooling across horizontal and vertical axes — and benchmark on the COVID-19 Radiography Dataset across normal, viral pneumonia, lung opacity, and COVID-19 classes with CLAHE contrast enhancement.

ConvNeXt-Nano CoordAtt 4-Class Classification PyTorch CLAHE Preprocessing

VIEW ON IEEE XPLORE

Interactive Pathology & Coordinate Attention Simulator

Toggle preprocessing stages and directional attention projections evaluated on 21,165 chest radiography scans.

COVID-19 RADIOGRAPHY DATASET

CLAHE CONTRAST ENHANCED (Optimized Ribcage Boundaries)

Architecture Backbone	Attention	Accuracy	Precision	Latency
ConvNeXt-Nano (Ours)	1D Coord Attn	98.4%	98.1%	14.2ms
Vision Transformer (ViT-B/16)	Self-Attention	95.2%	94.8%	41.5ms
DenseNet-121	None	94.6%	94.1%	29.1ms
ResNet-50 (Baseline)	None	93.1%	92.5%	22.8ms

* Evaluated across 4 classes: Normal, Viral Pneumonia, Lung Opacity, and COVID-19.