The identification of these autofluorescent substances sets the stage for the development of a novel imaging technique known as tissue autofluorescence NIR-II imaging (TANI). Unlike conventional imaging modalities, TANI operates without the need for exogenous contrast agents, thus promising to enhance the safety and efficiency of liver surgeries. This non-invasive method harnesses the inherent optical properties of tissues, providing surgeons with a powerful tool to improve intraoperative decision-making and outcomes.

Testing the capabilities of TANI revealed extraordinary contrast ratios, averaging at an impressive 7.69 ± 0.52. This remarkable contrast underscores TANI’s potential to effectively distinguish between various types of liver tumors, including hepatocellular carcinoma, intrahepatic cholangiocarcinoma, and metastatic lesions, regardless of the patient’s underlying liver condition, be it cirrhotic or non-cirrhotic. The high sensitivity of the imaging method, pegged at 97.8%, and an equally striking specificity rate of 98.4% showcase TANI’s robust performance in real clinical scenarios.

In contrast to existing techniques that utilize fluorescence-guided surgery in the visible light spectrum or the first near-infrared window, TANI demonstrates a superior ability to delineate cancerous tissues with minimal interference from benign lesions, blood, or bile contaminants. This is particularly noteworthy as the presence of such contaminants often complicates surgical procedures and leads to diagnostic challenges. The findings suggest a paradigm shift in how surgeons will approach liver surgeries, providing them with critical information that could alter the course of treatment on the spot.

Another significant advantage of TANI is its resilience against variations in cancer grade or stage. This consistency is essential as it underscores the technique’s reliability across a spectrum of liver malignancies, making it a valuable asset in the toolkit of contemporary surgical practices. Furthermore, the imaging capability is expected to enhance the surgeon’s ability to carry out oncological resections with greater precision, thereby reducing the likelihood of tumor recurrence post-surgery.

What truly elevates TANI as a groundbreaking application is its label-free nature. Surgeons can now rely on real-time imaging without administering any additional contrast agents, which could pose an extra risk to patients. This attribute aligns perfectly with the evolving focus towards patient-centric and minimally invasive surgical techniques that prioritize safety and effective patient outcomes.

The innovation presented by TANI not only holds promise for liver surgeries but may also extend its utility in other domains of oncology. The inherent autofluorescence properties observed may serve as a springboard for further research into other solid tumors. Indeed, the study hints at a strong correlation between near-infrared autofluorescence and various malignancies, inviting future inquiries into its applications across different tissue types and cancer forms.

Moreover, the opportunity to uncover molecular insights behind these autofluorescent substances could herald new diagnostic avenues. Understanding the biochemical nature of these substances may unlock additional layers of data crucial for distinguishing diseased tissues from healthy ones, possibly paving the path for the development of targeted therapies or new imaging biomarkers.

The promise of TANI as a revolutionary intraoperative management tool for liver cancer adds an exciting chapter to the ongoing pursuit of better diagnostic and therapeutic processes in oncology. As the medical community increasingly desires quick and reliable methodologies that lead to enhanced surgical performance, TANI stands as a testament to what innovative imaging technology can achieve.

With TANI proving its efficacy, the expectation is that implications will stretch far beyond the operating room, potentially influencing preoperative planning and overall cancer treatment strategies. When seamlessly integrated into surgical workflows, this technology may alter how liver malignancies are approached both from the surgical and the oncological perspective.

As researchers continue to explore the capabilities and expand the applications of TANI, one can anticipate a ripple effect across various medical specialties, leading to improvements in surgical techniques, enhanced patient safety, and ultimately, better clinical outcomes for individuals battling liver cancer.

To conclude, the advancements introduced through tissue autofluorescence NIR-II imaging represent a notable leap forward in the pursuit of precision medicine. As surgical practices continue to innovate and integrate groundbreaking technologies, the potential of TANI to reshape surgical oncology cannot be underscored enough. With promising clinical performance and strong foundational science, TANI is poised to set new standards in tumor visualization.

Subject of Research: Development of tissue autofluorescence NIR-II imaging (TANI) for visualization of human liver malignancies.

Article Title: Label-free tissue NIR-II autofluorescence imaging for visualization of human liver malignancy.

Article References:

He, H., Zhu, W., Miao, H. et al. Label-free tissue NIR-II autofluorescence imaging for visualization of human liver malignancy.
Nat. Biomed. Eng (2026). https://doi.org/10.1038/s41551-025-01593-4

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41551-025-01593-4

Keywords: liver malignancies, TANI, NIR-II imaging, surgical oncology, autofluorescence, intraoperative imaging, precision medicine, tumor visualization, cancer diagnosis, real-time imaging.

Prostate cancer remains one of the most prevalent malignancies among men worldwide, and its progression often leads to bone metastases—a critical factor influencing patient prognosis and treatment strategy. Conventional detection methods heavily rely on combined imaging modalities such as [^18F]FDG and [^18F]PSMA-1007 PET/CT scans. While effective in visualizing metastatic lesions, these methods are associated with significant drawbacks, including high operational costs and increased radiation exposure to patients. Addressing these limitations, the research team explored deep learning methods to synthesize functional PET images using only structural CT data, thereby promising a non-invasive, cost-effective alternative.

The study amassed a robust dataset comprising paired whole-body [^18F]PSMA-1007 PET/CT images from 152 subjects, carefully curated through retrospective analysis. These included 123 patients clinically and pathologically diagnosed with prostate cancer and 29 with benign lesions serving as comparative controls. The mean patient age was 67.48 years, with an average lesion size of approximately 8.76 millimeters. Such comprehensive data enabled the research to construct detailed bone structure images by preprocessing and segmenting both low-dose CT and PET scans, a crucial step for effective model training.

Central to the methodology was the deployment of two distinct GAN architectures: Pix2pix and CycleGAN. Both models are renowned for their capabilities in image-to-image translation tasks, but they approach the synthesis differently. Pix2pix operates on paired datasets with supervised learning, while CycleGAN leverages unpaired data through cycle consistency to achieve transformation. By training these networks to convert CT bone images into synthetic [^18F]PSMA-1007 PET images, the study rigorously assessed performance across multiple quantitative metrics including mean absolute error (MAE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index metric (SSIM), and importantly, the target-to-background ratio (TBR) relevant for identifying metastatic lesions.

Results from this extensive validation imparted compelling evidence of model efficacy. The Pix2pix model outperformed CycleGAN, attaining an exceptional SSIM of 0.97, indicative of near-perfect structural similarity between synthetic and real PET images. Additionally, a PSNR of 44.96 and low error rates (MSE at 0.80 and MAE at 0.10) underscored the precision of synthetic image generation. Particularly significant was the strong correlation (Pearson’s r > 0.90) observed between TBR values calculated from synthesized versus actual PET bone images—this parameter being critical for differentiating malignant bone lesions from healthy tissue with statistical insignificance in difference (p < 0.05). Such findings substantiate the concept that deep learning-generated synthetic PET images can reliably replicate the diagnostic information traditionally obtained from resource-intensive PET imaging. By effectively transforming routine low-dose CT images into functional molecular imaging maps, this approach portends a paradigm shift in oncological imaging workflows, making early detection of prostate cancer bone metastases more accessible and safer for patients. Beyond clinical implications, this technology also aligns with ongoing global efforts to reduce healthcare costs and patient radiation burden. Since PET imaging involves radioactive tracers and specialized equipment, its widespread use is often constrained by expense and availability. Synthetic imaging through GANs could democratize access to advanced diagnostics by harnessing the ubiquity of conventional CT scanners, which are less costly and more widely distributed across medical settings. The study’s pilot nature highlights the necessity for further multicentric clinical trials and larger datasets to optimize model generalizability across diverse patient populations and imaging protocols. Nonetheless, the promising preliminary outcomes lay the groundwork for integrating artificial intelligence seamlessly into clinical radiology, complementing rather than replacing traditional imaging. Scientifically, this research bridges the gap between anatomical and functional imaging through artificial intelligence. While CT provides detailed bone morphology, PET offers insight into metabolic activity relevant for cancer diagnosis and staging. Synthesizing these two imaging domains via GANs enables clinicians to infer molecular behavior from structural data, expanding the diagnostic utility of existing imaging resources without additional patient risk. Technically, the deployment of Pix2pix and CycleGAN GANs demonstrates the versatility of conditional adversarial networks in medical imaging. The Pix2pix’s utilization of paired datasets yields superior fidelity, but CycleGAN’s capacity for unpaired data remains advantageous for scenarios where such alignment is challenging. Future improvements may include model refinement, incorporation of 3D volumetric analysis, and fusion with clinical variables to enhance diagnostic accuracy. Moreover, the ability to accurately calculate TBR in synthetic images is vital, as this ratio is widely used to quantify lesion uptake relative to surrounding tissue. Maintaining statistical equivalence with real PET scans ensures clinical confidence in synthetic outputs, crucial for determining treatment response and prognosis in prostate cancer patients. In conclusion, this pilot validation study illustrates a significant leap towards AI-driven synthetic molecular imaging, opening avenues for safer, economical, and widely accessible cancer diagnostics. By synthesizing [^18F]PSMA-1007 PET bone images from low-dose CT, deep learning models promise to reduce unnecessary radiation, lower healthcare costs, and expedite early detection of bone metastases in prostate cancer, ultimately enhancing patient outcomes and quality of life. As artificial intelligence continues to evolve, its integration with radiologic imaging heralds a new frontier in precision medicine. The convergence of advanced machine learning algorithms with routine imaging modalities may soon redefine standard diagnostic pathways, enabling earlier interventions and personalized therapeutic strategies. Continued interdisciplinary collaboration between oncologists, radiologists, and AI specialists will be vital to translate these promising findings into clinical practice. This research marks an exciting milestone in leveraging computational power to augment human expertise and transform oncologic imaging. The potential to synthesize intricate molecular data from conventional scans reshapes our understanding of diagnostic imaging, setting the stage for innovations that prioritize patient safety, accessibility, and accuracy. Subject of Research: Early detection of prostate cancer bone metastases using synthetic [^18F]PSMA-1007 PET images generated from CT scans by deep learning techniques. Article Title: Synthesizing [^18F]PSMA-1007 PET bone images from CT images with GAN for early detection of prostate cancer bone metastases: a pilot validation study. Article References: Chai, L., Yao, X., Yang, X. et al. Synthesizing [^18F]PSMA-1007 PET bone images from CT images with GAN for early detection of prostate cancer bone metastases: a pilot validation study. BMC Cancer 25, 907 (2025). https://doi.org/10.1186/s12885-025-14301-x Image Credits: Scienmag.com DOI: https://doi.org/10.1186/s12885-025-14301-x