|
|
 |
| |
|
|
|
Agent-Based Intelligent Systems for Pneumonia Detection and Explainable Diagnostic Reporting: A Review |
|
|
|
PP: 1269-1278 |
|
|
doi:10.18576/amis/200512
|
|
|
|
Author(s) |
|
|
|
Esam Othman,
Fadl Dahan,
|
|
|
|
Abstract |
|
|
| Pneumonia remains one of the leading infectious causes of death worldwide, particularly among children under five and older adults; recent global burden estimates attribute approximately 2.5 million deaths in 2023 alone to the disease, including more than 600,000 children under five. Chest X-ray (CXR) imaging is the most widely available diagnostic modality, yet its interpretation remains subject to inter-reader variability and is constrained by the limited availability of specialist radiologists in many regions. Deep learning, and convolutional neural networks (CNNs) in particular, has driven a major shift in the automated detection of pneumonia from chest radiographs, beginning with seminal works such as CheXNet and the study of Kermany et al., and progressing through transfer-learning architectures including VGG, ResNet, DenseNet and EfficientNet. Nevertheless, most such models remain largely opaque “black boxes” that lack the transparency required to gain clinician trust, motivating explainable AI (XAI) techniques such as Grad-CAM, LIME and SHAP. More recently, multi-agent systems built on large language models (LLMs) that emulate the collaboration of multidisciplinary medical teams – exemplified by MedAgents, MDAgents, AgentClinic and KG4Diagnosis – have emerged as a promising direction for integrating visual perception, knowledge-based reasoning and report generation into a single explainable diagnostic pipeline. This paper presents a systematic literature review, guided by the PRISMA framework, of 32 studies covering (i) deep learning models for pneumonia detection, (ii) explainable AI techniques in medical imaging, and (iii) multi-agent systems for medical diagnosis and report generation, complemented by comparative tables of relevant studies and datasets and by a proposed conceptual architecture for an integrated agent-based system that combines diagnostic accuracy with clinical transparency. By supporting earlier and more accessible diagnosis of a leading cause of preventable death, this line of research also contributes to Sustainable Development Goal 3 (Good Health and Well-Being). The review concludes by discussing open challenges and future research directions in this emerging field. |
|
|
|
|
|
|
|
