JMIR Publications

ABSTRACT

Background:

Explainable artificial intelligence (XAI) is increasingly applied in biomedical imaging to address the demand for transparency, clinical interpretability, and trust in AI-driven decision support systems, particularly in cancer care.

Objective:

This scoping review maps current research at the intersection of explainable artificial intelligence (XAI) and biomedical imaging for cancer care.

Methods:

We conducted a structured search in PubMed and Scopus, selecting 171 studies published from 2017 to 2023. Inclusion criteria focused on peer-reviewed research applying AI/ML methods to radiological imaging data with an explicit emphasis on explainability or interpretability.

Results:

Deep learning, particularly convolutional neural networks (CNNs), dominated methodological choices, with post-hoc techniques such as Grad-CAM and SHAP widely used. Only a small proportion of studies used inherently transparent models. Explainability was primarily applied to improve feature traceability, clinical understanding, and user trust. However, only a minority of studies conducted formal validation of the interpretability component. Imaging modalities spanned CT, MRI, ultrasound, and mammography, with breast and lung cancers most frequently addressed. Most studies relied on private datasets, and few were integrated into decision support systems.

Conclusions:

While XAI methods are frequently used in cancer imaging, challenges remain in standardization, interpretability validation, and clinical integration. This review identifies critical trends and gaps in the use of explainable AI in oncological imaging and highlights opportunities for enhancing model transparency and usability in real-world medical settings.