A Novel Multi-branch ConvNeXt Architecture for Identifying Subtle Pathological Features in CT Scans
Irash Perera, Uthayasanker Thayasivam
TL;DR
The paper tackles the challenge of diagnosing subtle pathological features in CT scans, using COVID-19 as the representative task. It introduces a novel multi-branch ConvNeXt architecture that fuses global, salient, and attention-weighted features, built on a robust end-to-end pipeline including CLAHE-based enhancement, lung ROI cropping, and balanced data augmentation. A two-phase transfer-learning strategy—initially training a frozen base, then fine-tuning part of the base—yields a ROC-AUC of 0.9937 and an F1-score of 0.9825 on a 701-slice validation set, outperforming several state-of-the-art baselines on the same datasets. The results demonstrate the approach’s high discriminative power and practical potential for robust medical diagnostics, with broader applicability to other CT-based pathology classification tasks.
Abstract
Intelligent analysis of medical imaging plays a crucial role in assisting clinical diagnosis, especially for identifying subtle pathological features. This paper introduces a novel multi-branch ConvNeXt architecture designed specifically for the nuanced challenges of medical image analysis. While applied here to the specific problem of COVID-19 diagnosis, the methodology offers a generalizable framework for classifying a wide range of pathologies from CT scans. The proposed model incorporates a rigorous end-to-end pipeline, from meticulous data preprocessing and augmentation to a disciplined two-phase training strategy that leverages transfer learning effectively. The architecture uniquely integrates features extracted from three parallel branches: Global Average Pooling, Global Max Pooling, and a new Attention-weighted Pooling mechanism. The model was trained and validated on a combined dataset of 2,609 CT slices derived from two distinct datasets. Experimental results demonstrate a superior performance on the validation set, achieving a final ROC-AUC of 0.9937, a validation accuracy of 0.9757, and an F1-score of 0.9825 for COVID-19 cases, outperforming all previously reported models on this dataset. These findings indicate that a modern, multi-branch architecture, coupled with careful data handling, can achieve performance comparable to or exceeding contemporary state-of-the-art models, thereby proving the efficacy of advanced deep learning techniques for robust medical diagnostics.