Lightening tumors: AI-powered fluorescence imaging enables a new era of precise diagnosis and treatment
image: A complete technological chain for fluorescence molecular imaging, ranging from imaging instrumentation, imaging principles, and near-infrared fluorescent probe design to multimodal imaging, theranostics integration, and intelligent analysis.
Credit: Composite figure by Bingrui Wang, Jiaxuan Wen and Yang Du, Institute of Automation, Chinese Academy of Sciences. A portion was adapted from Z. Zhang et al., “NIR-II light in clinical oncology: opportunities and challenges,” Nature Reviews Clinical Oncology, vol. 21, pp. 449–467, 2024, © Springer Nature Limited 2024, used with permission. Portions were adapted from X. Zhu et al., “Prediction of lymph node metastasis in colorectal cancer using intraoperative fluorescence multi-modal imaging,” IEEE Transactions on Medical Imaging, vol. 44, no. 3, pp. 1568–1580, 2025, licensed under CC BY 4.0; and B. Shen et al., “Real-time intraoperative glioma diagnosis using fluorescence imaging and deep convolutional neural networks,” European Journal of Nuclear Medicine and Molecular Imaging, vol. 48, no. 11, pp. 3482–3492, 2021, licensed under CC BY 4.0. Changes were made.
This review mainly discusses two aspects—probe innovation and AI empowerment—to summarize recent research progress in fluorescence molecular imaging (FMI) for tumor diagnosis, therapy, and treatment-response assessment. It further outlines key technical pathways and discusses major barriers to clinical translation.
- Functional fluorescence imaging of tumors based on probes
FMI has evolved from early nonspecific probes that relied on the EPR effect to intelligent probe systems endowed with active targeting and tumor microenvironment responsiveness. Active targeting enables tumor enrichment through antibody/peptide recognition or cell-membrane biomimicry. Responsive probes use tumor characteristic signals (e.g., enzyme overexpression, acidic pH, hypoxia, etc.) to achieve “on-demand activation,” thereby improving the signal contrast between tumors and background tissues. Meanwhile, the NIR-II imaging window reduces tissue scattering and absorption and markedly attenuates autofluorescence, enhancing visualization of deep-seated tumors and small-volume lesions. Furthermore, multimodal probes and theranostic integrated probes couple imaging functions with therapeutic capabilities, driving FMI from a single imaging modality toward a clinical diagnostic-therapeutic platform.
2. AI-Enhanced Fluorescence Imaging and 3D Reconstruction
At the imaging level, AI improves fluorescence image quality through deep learning, enhancing signal-to-noise ratio and spatial resolution, which provides more reliable visual evidence for intraoperative tumor boundary delineation. For 3D information acquisition, AI is also applied to reconstruction tasks such as fluorescence molecular tomography. Learning-based approaches can mitigate the “ill-posedness” and computational slowness of conventional algorithms, enabling faster and more accurate reconstruction of 3D probe distributions from 2D observations and improving the robustness of localization and reconstruction.
3. Clinical Advances and Barriers in AI-Enabled Functional Fluorescence Imaging
In clinical applications, AI-enabled FMI has shown promising performance in tasks including intraoperative tumor grading, lymph node metastasis prediction, and margin/boundary assessment. By leveraging multimodal or multispectral fusion, it enhances quantitative diagnostic capability and supports real-time decision-making. Finally, the review objectively evaluates current barriers to clinical translation, including limited clinical data and insufficient standardization, concerns regarding the interpretability and trustworthiness of deep learning models, the need to further improve algorithmic speed and device integration capability, and the lack of unified guidelines and multicenter validation. Overall, the synergistic development of probe innovation and AI optimization is pushing FMI toward a future of greater intelligence, precision, and integration in diagnosis and therapy.