Review Article

 Advances in artificial intelligence and new imaging technologies in clinical trials: A Mini-Review

Wooseok Jeon¹

Abstract

Background/Objectives: Artificial intelligence (AI) is rapidly transitioning from a supportive tool into a core driver of innovation across the field of medical imaging. With the growing complexity of clinical data and imaging modalities, AI-based approaches have the potential to significantly enhance diagnostic accuracy and support personalized treatment strategies. This review aims to provide an integrated overview of AI applications in medical imaging and clinical trials, with a particular focus on precision medicine.

Methods: A comprehensive literature search was conducted using PubMed and Google Scholar to identify relevant studies published up to June 30, 2025. Search terms included "artificial intelligence," "medical imaging," "precision medicine," "clinical trials," and "multimodal learning." Articles were screened to evaluate AI-driven solutions involving clinical, biological, and imaging data across oncology and other diseases.

Results: Multimodal machine learning techniques that integrate clinical, biological, and imaging datasets have shown great promise in improving diagnostic performance and patient stratification. AI-enhanced imaging and robotics technologies have demonstrated the ability to increase diagnostic standardization and accelerate clinical translation. Furthermore, AI is helping to bridge diagnostic and therapeutic domains, contributing to the development of a fully integrated precision medicine ecosystem.

Conclusions: AI-based technologies are expected to transcend the traditional boundaries of medical imaging by fostering the convergence of diagnosis and treatment. These innovations support the clinical realization of precision medicine by enhancing decision-making, accelerating clinical workflows, and enabling personalized treatments.

Keywords

Artificial intelligence, Medical imaging, Precision medicine, Clinical trials, Multimodal learning

Introduction

Molecular imaging technologies (PET, MRI, SPECT) have emerged as key tools for the early diagnosis and precise pathophysiological assessment of diseases, forming the foundation of precision medicine. PET enables metabolic and receptor-based imaging, MRI provides high-resolution anatomical and functional imaging, and SPECT offers pathological specificity, playing critical roles in oncology, neurology, and cardiovascular disease management.

Recently, artificial intelligence (AI) and deep learning-based imaging analysis have become indispensable for the quantification, reproducibility, and standardization of these modalities. AI supports lesion detection, segmentation, and quantification while integrating clinical data for objective decision support. In parallel, novel radiotracers (e.g., 68Ga-FAPI, 18F-GP1), nanoparticle-based contrast agents (e.g., Ferumoxytol), and image-guided robotic surgery are being validated in clinical trials, bridging gaps in conventional imaging and accelerating the convergence of precision diagnosis and therapy.

This review aims to comprehensively evaluate the clinical value and applicability of AI, deep learning-based image correction, novel PET tracers, and image-guided robotics based on recent clinical trials. Furthermore, it discusses their role in advancing quantification, automation, and standardization in medical imaging, outlining future directions for their integration into precision medicine.

 AI-Based Imaging Analysis AI in PET/CT

In breast cancer, Miller R et al. developed a convolutional neural network (CNN)-based deep learning model to automatically detect metastatic lesions on 18F-FES PET/CT scans in 52 patients. The model achieved a sensitivity of 62% and improved inter-reader concordance, demonstrating the feasibility of automated lesion classification. These findings highlight the potential of AI-driven image analysis to standardize interpretation and reduce inter-observer variability in metastatic assessment [1]. In lung cancer, the research team of  Caiyue Ren et. al,  analyzed 260 patients and developed a clinico-biological-radiomics (CBR)-based machine learning model by integrating clinical data, biological features, and FDG-PET radiomics. The model achieved an AUC of 0.90 and reduced the false-positive rate (FPR) to 6.45%, significantly enhancing the precision of lymph node assessment. This approach suggests that AI-driven multimodal integration may reduce unnecessary invasive procedures and support the development of personalized treatment strategies [2]. Furthermore, the research team of Christos Sachpekidis et, al, investigated 35 patients with multiple myeloma and developed a deep learning-based PET quantification model. This model enabled automated calculation of metabolic tumor volume (MTV) and total lesion glycolysis (TLG), which showed significant correlations with pathological bone marrow infiltration and β2-microglobulin levels. These results demonstrated superior reproducibility compared to manual quantification and highlighted the potential of AI-based PET analysis as a standardized tool for disease burden assessment and treatment response monitoring [3]. Collectively, these studies underscore the emerging role of AI-driven PET imaging in improving diagnostic accuracy, ensuring reproducibility, and providing quantitative insights that can inform precision oncology across different cancer types.

AI in Image Segmentation

The research team of Mahsa Torkaman et. al, developed an organ-focused deep learning model for PET segmentation in gastrointestinal tumors. Compared to a conventional whole-body segmentation approach, this organ-specific model demonstrated a significantly higher Dice similarity coefficient (0.78 vs. 0.63, p<0.0001), indicating superior accuracy in delineating tumor boundaries. This result highlights the importance of organ-focused AI models in improving segmentation performance, particularly for anatomically complex regions where whole-body models may lack precision.

In addition, recent advancements in AI-driven diagnostic performance have shown that artificial intelligence not only enhances lesion detection and segmentation but also improves the accuracy of quantitative metrics such as metabolic tumor volume (MTV) and total lesion glycolysis (TLG). Furthermore, AI reduces false-positive findings, ultimately supporting its role as a valuable tool for clinical decision-making and treatment response assessment. These findings collectively underscore the transformative potential of AI in refining PET imaging workflows and enabling more precise and reproducible evaluations in oncologic imaging [4].

Deep Learning-Based Image Correction

The research team of Telma Sprauel et. al, investigated the use of deep learning–based attenuation correction (DLAC) in SPECT imaging for 84 patients at risk of coronary artery disease (CAD). The DLAC-SPECT method significantly improved right coronary artery (RCA) uptake uniformity (81.2% vs. 69.1% with standard SPECT) and reduced stress-rest fluctuations, demonstrating a performance that closely matched rubidium PET (Rb-PET). These findings suggest that DLAC can enhance diagnostic accuracy in myocardial perfusion imaging by improving image consistency and reducing the need for PET in certain clinical settings [5]. 

The research team of Hidenobu Hashimoto et. al, compared silicon photomultiplier (SiPM) PET with conventional photomultiplier tube (PMT) PET in 25 patients. SiPM PET markedly reduced image noise (4.0% vs. 7.6%), improved signal-to-noise ratio (SNR) (32.7 vs. 16.3), and increased target-to-background ratio (TBR) (1.1 vs. 0.8), thereby enabling superior coronary plaque imaging quality. These results highlight the technological advantage of SiPM PET over PMT PET in providing clearer and more reliable cardiac imaging, which could translate into improved clinical assessment of coronary artery disease [6].

Novel Tracers and Advanced Imaging

The research team of Hui Yuancompared 68Ga-FAPI PET with FDG PET in 26 patients with solid tumors. The study demonstrated that FAPI PET achieved a markedly higher sensitivity for detecting abdominal metastases (1.0 vs. 0.475 for FDG), highlighting its potential to improve metastatic disease evaluation. Furthermore, FAPI PET was noted to have promising theranostic applications by targeting fibroblast activation protein, which may allow both diagnostic imaging and targeted radioligand therapy in the future [7].

Beth Whittington et. al, the research team evaluated 18F-GP1 PET imaging in 11 patients with ischemic stroke or transient ischemic attack (TIA). 18F-GP1 PET successfully identified thrombi in 10 out of 11 patients and showed strong concordance with CT findings. These results not only support the feasibility of using 18F-GP1 PET for in vivo thrombus detection but also provide valuable insights into the pathophysiology of thrombosis, which may inform future therapeutic strategies [8].

The research team of Ali Rashidi et.al, investigated Ferumoxytol-enhanced MRI in 26 pediatric cancer patients to improve the detection of bone marrow metastases. Compared to unenhanced MRI, Ferumoxytol-MRI significantly increased sensitivity (96% vs. 83%) and diagnostic accuracy (99% vs. 95%). This study underscores the value of Ferumoxytol as an iron-based contrast agent in enhancing bone marrow lesion visualization and improving diagnostic confidence in pediatric oncology [9].

Image-Guided Robotics

The research team of Yichen Xu evaluated the use of neuro-robotic navigation in 52 patients undergoing epilepsy surgery, with a particular focus on MRI-negative cases where lesion localization is challenging. By integrating multimodal imaging fusion into the neuro-robotic platform, the study significantly improved surgical accuracy, resulting in a markedly higher rate of favorable outcomes (ENGEL I: 88.2% vs. 50% with conventional approaches). These findings underscore the clinical value of combining advanced imaging techniques with robotic assistance to enhance surgical precision and improve postoperative seizure control in complex epilepsy cases [10].

Table 1. Key imaging challenges and corresponding solutions with clinical evidence

Discussion

Traditional imaging modalities such as PET, SPECT, and MRI have provided valuable insights into disease diagnosis and monitoring but remain limited by several fundamental challenges, including interobserver variability, lack of standardized quantification, insufficient disease-specific tracers, and limited integration with image-guided interventions. Recent advances in artificial intelligence (AI), deep learning–based image correction, novel molecular tracers, and image-guided robotics are now addressing these limitations, offering strong clinical evidence for their transformative potential in precision medicine These findings demonstrate that emerging technologies do more than provide incremental improvements; they redefine the standard of care by directly addressing long-standing diagnostic limitations.

AI-based lesion detection and quantification reduce observer variability and enable reproducible, objective evaluation of disease burden, which is critical for patient stratification and treatment monitoring. Deep learning–based image correction and SiPM PET have elevated image quality and quantitative reliability to levels that approach or even exceed current standards, particularly in cardiovascular imaging. Meanwhile, novel tracers such as 68Ga-FAPI, 18F-GP1, and Ferumoxytol-MRI have expanded the scope of molecular imaging to areas previously inaccessible to FDG-PET, including tumor stroma, thrombi, and bone marrow metastases. Furthermore, the integration of multimodal imaging into robotic-assisted surgery has demonstrated tangible clinical benefits by improving surgical accuracy in complex neurological procedures such as MRI-negative epilepsy.

Despite these advances, challenges remain. Large-scale multicenter validation studies are needed to establish standardized protocols for AI-driven quantification and automation. Similarly, prospective clinical trials will be essential to confirm the clinical utility of novel tracers and to integrate molecular imaging into image-guided robotic platforms. The development of international guidelines for AI-based imaging analytics, tracer standardization, and imaging-robotics integration will be crucial for translating these technologies from early clinical trials into routine clinical practice.

Conclusion

Collectively, these advancements mark a critical step toward the realization of precision medicine, where AI-driven analytics, advanced molecular imaging, and image-guided robotics converge to deliver highly individualized diagnosis and therapy. AI, deep learning, novel imaging tracers, and image-guided robotics are rapidly transforming the landscape of precision diagnosis and therapy. Emerging clinical evidence demonstrates their significant value in enhancing diagnostic accuracy, improving quantitative imaging reliability, and enabling unprecedented surgical precision. As these technologies continue to mature and undergo broader clinical validation, they are poised to become the cornerstone for seamlessly integrating precision medicine into routine clinical practice, ultimately driving more personalized and effective patient care.

Conflict of Interest

The authors have no conflicts of interest to declare and agreed to the published version of the manuscript.

Author Contributions

References

1. Miller R, Battle M, Wangerin K, Huff DT, Weisman AJ, Chen S, Perk TG, Ulaner GA. Evaluating Automated Tools for Lesion Detection on 18F Fluoroestradiol PET/CT Images and Assessment of Concordance with Standard-of-Care Imaging in Metastatic Breast Cancer. Radiol Imaging Cancer. 2025;7(3):e240253.
2. Ren C, Zhang F, Zhang J, Song S, Sun Y, Cheng J. Clinico-biological-radiomics (CBR) based machine learning for improving the diagnostic accuracy of FDG-PET false-positive lymph nodes in lung cancer. Eur J Med Res. 2023;28(1):554.
3. Sachpekidis C, Enqvist O, Ulén J, Kopp-Schneider A, Pan L, Jauch A, Hajiyianni M, John L, Weinhold N, Sauer S, Goldschmidt H, Edenbrandt L, Dimitrakopoulou-Strauss A. Application of an artificial intelligence-based tool in [18F]FDG PET/CT for the assessment of bone marrow involvement in multiple myeloma. Eur J Nucl Med Mol Imaging. 2023;50(12):3697–3708.
4. Torkaman M, Jemaa S, Fredrickson J, Fernandez Coimbra A, De Crespigny A, Carano RAD. Comparative analysis of intestinal tumor segmentation in PET CT scans using organ based and whole body deep learning. BMC Med Imaging. 2025;25(1):52.
5. Sprauel T, Imbert L, Doyeux K, Perrin M, Claudin M, Roch V, Doyen M, Roth N, Lamash Y, Chequer R, Rouzet F, Hyafil F, Marie PY. Assessment of sestamibi CZT-SPECT reconstructed using deep-learning-based virtual attenuation correction maps according to coronary artery territory and with comparison to rubidium-PET. J Nucl Cardiol. 2025;49:102226.
6. Hashimoto H, Kuronuma K, Hyun MC, Han D, Builoff V, Cadet S, Dey D, Berman DS, Kwiecinski J, Slomka PJ. Head-to-head comparison of 18F-sodium fluoride coronary PET imaging between a silicon photomultiplier with digital photon counting and conventional scanners. J Nucl Cardiol. 2024;42:102045.
7. Yuan H, Liu E, Zhang G, Lai C, Zhang Q, Shang Y, Cheng Z, Jiang L. Diagnostic efficacy of [68Ga]Ga-DOTA-GPFAPI-04 in patients with solid tumors in a head-to-head comparison with [18F]F-FDG: results from a prospective clinical study. Eur J Nucl Med Mol Imaging. 2024;51(11):3360–3372.
8. Whittington B, Tzolos E, Bing R, Andrews J, Lucatelli C, MacAskill MG, Tavares AAS, Clark T, Mills NL, Nash J, Dey D, Slomka PJ, Koglin N, Stephens AW, van Beek EJR, Smith C, Dweck MR, Williams MC, Whiteley W, Wardlaw JM, Newby DE. Noninvasive In Vivo Thrombus Imaging in Patients With Ischemic Stroke or Transient Ischemic Attack-Brief Report. Arterioscler Thromb Vasc Biol. 2023;43(9):1729–1736.
9. Rashidi A, Baratto L, Theruvath AJ, Greene EB, Jayapal P, Hawk KE, Lu R, Seekins J, Spunt SL, Pribnow A, Daldrup-Link HE. Improved Detection of Bone Metastases in Children and Young Adults with Ferumoxytol-enhanced MRI. Radiol Imaging Cancer. 2023;5(2):e220080.
10. Xu Y, Chen Y, Liu H, Zhang H, Yin Z, Liu D, Zhu G, Diao Y, Wu D, Xie H, Hu W, Zhang X, Shao X, Zhang K, Zhang J, Yang A. The clinical application of neuro-robot in the resection of epileptic foci: a novel method assisting epilepsy surgery. J Robot Surg. 2023;17(5):2259–2269.