Introduction

Cervical cancer, the fourth most common cancer in women worldwide,¹ currently requires a 3 to 6-month waiting period for post-treatment PET-CT assessment.² Earlier treatment-response evaluation would make it possible to alter treatment for women with poor response to initial therapy.

Restriction spectrum imaging (RSI) has demonstrated potential in cervical cancer diagnosis and monitoring.³ We have improved a cervical cancer-specific RSI model by 1) expanding our patient database, 2) using RSI data from healthy cervixes to create Z-score maps in cancer patients, and 3) demonstrating the effectiveness of RSI Z-score maps in distinguishing cancer from healthy tissue without the need for contrast agents.

Methods

Diagnostic MRI at 3T was performed on 15 cervical cancer patients with confirmed pathology, prior to treatment (average age: 46.5±14.6 years-old) and on 11 healthy volunteers (average age: 43.6±14.8 years-old). DWI pulse sequence parameters are provided in Table 1. These datasets were used convert cancer data to RSI Z-score maps. Additional 11 volunteers (average age: 44.5±12.4 years-old) were included to compare RSI Z-score results in cancer patients with an independent negative control group.

Cervix ROIs were drawn on DWI images from healthy controls and cancer ROIs were drawn on images from cancer patients by a fellowship-trained body radiologist. Conventional ADC maps were computed.

In the cervical cancer-specific RSI model, global ADCs are first estimated across all voxels from cancer ROIs using multi-exponential models. Once the fixed ADCs are determined, the signal contributions, C_i,N, are estimated. This approach allows the use of a consistent model across all voxels for comparing signal contributions across water compartments. Additionally, DWI data were noise corrected and averaged across diffusion directions for each b-value. The diffusion signal was then modeled as the linear combination of multiple exponential decays:

$$$S_{diff}(b,N) = S_o\sum_{k=1}^ne^{-b D_{i,N}}= \sum_{k=1}^nC_{i,N}e^{-b D_{i,N}}$$$ (1)

where N is the total number of exponential decays (here 2, 3 or 4), C_i,N are the signal contributions of each exponential component, b are the b-values in s/mm², and D_i,N are the ADCs of each exponential component (D_1,N<D_i,N).

Signal contribution C_i,N maps of cancer patients were then converted into a Z-score based on the average and standard deviation of healthy volunteers:

$$$Z_{i,N} = \frac{C_{i,N} - \mu_{i,N-healthy}}{\sigma_{i,N-healthy}}$$$ (2)

where and are the mean and standard deviation of the RSI C_i,N (Equation 1) in healthy cervix. RSI Z-maps indicate deviations in RSI signal contribution from healthy tissue.

The relative Bayesian information criterion (BIC) was estimated for all multi-exponential models because it penalizes the number of parameters.4 Lower BIC values indicate improved model fitting. Signal contributions C_i,N and RSI Z-scores were compared between cancerous and healthy cervical tissues for the three RSI models using an unpaired two-tailed t-test, and a Bonferroni correction was used for multiple comparisons.

Results

Fixed ADCs and BIC values for each model are shown in Table 2. BIC scores indicate that bi-exponential models are sufficient for describing the diffusion-weighted signal of cervical cancer tissues. However, higher order RSI models remove signal from other tissues increasing tumor conspicuity (Figures 1-2). The RSI signal contribution maps from bi- and tri-exponential models were similar (Figures 1-2). The main difference was that the compartment C_3,3 appears to include signals from fast moving fluids (e.g., urine and vascular flow).

Conventional ADCs were different between cervical cancers and heathy cervixes. RSI signal contributions C_1,2, C_1,3, C_2,3, and C_2,4, were higher (p<0.05) in tumor tissues (Table 3). The compartment C_1,3 and C_2,4 exhibited the largest difference between cervical cancer and healthy cervixes (Group A). These results were also reflected on the RSI C_i,N Z-score maps of tumors when compared to those from the negative control (Table 3).

Discussion and Conclusions

We demonstrate the potential of cervical cancer-specific RSI models in distinguishing tumors and healthy tissues. Although ADC ROI values were statistically different between groups, RSI models enhanced tumor conspicuity without the need for exogenous contrast in women with cervical cancer ranging from stages IA2 to IIIC1. Future work will include assessing tumor conspicuity and the clinical utility of RSI in cervical cancer evaluation.

The RSI compartments associated with restricted and hindered diffusion (i=12) were different between cervical cancer and healthy cervixes. These compartments are attributed to cancer cells, while the faster compartments (i=3 or 4) may originate from tissues with free diffusion and vascular flow.

In future work, we will evaluate the diagnostic value of RSI models in an independent cohort of cancer patients compared to DCE-MRI and PET/MRI and assess the ability of the RSI model to identify post-treatment edema.

Acknowledgements

Supported by NIH R37CA249659 and a research grant from General Electric Healthcare.