Charged Gold Nanoparticles for Target Identification-Alignment and Automatic Segmentation of CT Image-Guided Adaptive Radiotherapy in Small Hepatocellular Carcinoma

Abstract

Because of the challenges posed by anatomical uncertainties and the low resolution of plain computed tomography (CT) scans, implementing adaptive radiotherapy (ART) for small hepatocellular carcinoma (sHCC) using artificial intelligence (AI) faces obstacles in tumor identification-alignment and automatic segmentation. The current study aims to improve sHCC imaging for ART using a gold nanoparticle (Au NP)-based CT contrast agent to enhance AI-driven automated image processing. The synthesized charged Au NPs demonstrated notable in vitro aggregation, low cytotoxicity, and minimal organ toxicity. Over time, an in situ sHCC mouse model was established for in vivo CT imaging at multiple time points. The enhanced CT images processed using 3D U-Net and 3D Trans U-Net AI models demonstrated high geometric and dosimetric accuracy. Therefore, charged Au NPs enable accurate and automatic sHCC segmentation in CT images using classical AI models, potentially addressing the technical challenges related to tumor identification, alignment, and automatic segmentation in CT-guided online ART.

Keywords: Adaptive radiotherapy; Computed tomography imaging; Gold nanoparticles; Hepatocellular carcinoma; Tumor automatic segmentation.

Figure 1

Flowchart of charged Au NPs for CT image-guided adaptive radiotherapy (ART) of small hepatocellular carcinoma (sHCC) to identify, align, and automatically segment tumor target areas. (A) The study involved a standard procedure in which the charged Au NP reagent was prepared and administered through tail vein injection in tumor-bearing mice. Later, charged Au NPs aggregated in the weakly acidic tumor environment, inducing a significant elevation in particle size and prolonging their retention time. Subsequent in vivo CT imaging experiments were performed. Later, the AI autosegmentation model helped automatically identify and segment the tumors on the CT images. (B) The mechanism of charged Au NP aggregation in the weakly acidic tumor environment. (C) AI Automatic-Segmentation Model (3D U-net) network structure.

Figure 2

Characterization of Au NPs and charged Au NPs. (A) Characterization of Au NPs and charged Au NPs. (B,C) TEM images of Au NPs and charged Au NPs under different pH conditions over time. (D) Hydrated particle size dispersion and column diagrams of charged Au NPs. (E) Time-dependent variation in hydration particle size of Au NPs and charged Au NPs. (F) Absorption of Au NPs and (G) charged Au NPs versus pH.

Figure 3

In vitro and in vivo biosafety evaluation of Au NPs and charged Au NPs. Viability of sHCC cells after (A) 24 and (B) 48 h incubation using Au NPs or charged Au NPs. (C) Serum concentrations of ALT, AST, ALB, BUN, and CREA. (D) Primary indicators of routine blood testing after treatment. (E) H&E staining of various tissues and organs.

Figure 4

In vivo CT imaging of mice after injection with Au NPs and charged Au NPs. (A) Imaging at 6, 24, and 48 h after injecting Au NPs (HCC is marked blue). (B) Box plot of tumor volume, tumor electron density (ED), and liver ED depending on CT images obtained from 11 mice injected with Au NPs. (C) Imaging at 0, 6, 24, 72, 96, 120, and 144 h after injection of charged Au NPs (HCC is marked blue). (D) Box plot of tumor volume, tumor ED, and liver ED in seven mice injected with charged Au NPs. “0 h” demonstrates the CT images captured before injection.

Figure 5

Performance of two artificial intelligence (AI) models in automated segmenting small hepatocellular carcinoma (sHCC) in CT images of mice injected with charged Au NPs. (A) The network architectural principles of two classic artificial intelligence automatic segmentation models, 3D U-Net and 3D Trans U-Net. 3D U-Net: Reproduced or adapted with permission from ref (55). Copyright 2015, Springer Nature. 3D Trans U-Net: Reproduced with permission under CC-BY 4.0 license from ref (56). Copyright 2021, The Authors. (B) 10-fold cross-validation schematic. (C) Bar chart (accuracy, sensitivity, specificity, and AUC) of 10-fold cross-validation results for both AI models on this task data set. (D) A diagram of the radiotherapy planning target volume (outer contours) generated through the expansion of the sHCC tumor volume (inner contours) by 5 mm. (E) A radiation oncologist segmentation case (blue contour), 3D U-net automatic segmentation (green contour), and 3D Trans U-Net automatic segmentation (red contour) of CT cross-sectional images. (F) Radiation beam angle and radiation dose distribution in radiotherapy planning. (G) Construction of a 3D stacked histogram (Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), and Union of Intersection (UoI)) according to the segmentation accuracy of the AI models. (H) Violin plot (V5000 cGy, CI) of experimental results for radiotherapy planning using tumor volume from the two types of automatic segmentation.