Neural Thoracic X-Ray Pathology Classifier with Grad-CAM Saliency

Multi-label pathology screening across 112,000 chest radiographs (ChestX-ray14 cohort). Incorporates hybrid ResNet-50 backbones connected with Vision Transformer (ViT) cross-attention layers, mapping localized visual justifications to mitigate black-box misdiagnoses.

category Domain Deep Learning & Vision Teesside AI Lab

calendar_month Timeline Mar 2024 — Nov 2024 Published May 2026

science Status

Completed

Peer Review Ready

analytics Key Metric Mean AUC-ROC: 0.961 Hybrid ResNet-ViT

insights

Empirical Benchmarks & Model Comparison

Tested Performance

Mean AUC-ROC

0.961

Hybrid ResNet-ViT 14 Chest Pathologies

psychology

Problem Formulation & Investigative Scope

Problem Statement

Radiologist workload shortages in low-resource emergency departments lead to delayed triage of acute conditions such as pneumothorax and pulmonary edema.

Research Questions (RQs)

RQ1: Can cross-attention between ResNet spatial features and ViT global patch tokens increase multi-label classification accuracy over standalone convolutional backbones?

Structured Objectives

1 Benchmark ResNet-50 versus ViT-Base on ChestX-ray14 multi-label screening.

2 Implement Grad-CAM saliency extraction with radiologist IoU validation.

Explore Related Investigations

Clinical AI & Healthcare

Multimodal Clinical Decision Support for Sepsis Onset Prediction

Constructing an end-to-end continuous risk trajectory engine for ICU patient encounters. Combines irregular time-series vital signs, high-frequency lab assays, and clinical notes to forecast septic shock up to 6 hours prior to clinical diagnostic criteria with localized FastSHAP explanations.

Read Investigation arrow_forward