Objectives
Assessment of empirical diffusion-weighted MRI (DW-MRI) models in cervical tumours to investigate whether fitted parameters distinguish between types and grades of tumours.
Methods
Forty-two patients (24 squamous cell carcinomas, 14 well/moderately differentiated, 10 poorly differentiated; 15 adenocarcinomas, 13 well/moderately differentiated, two poorly differentiated; three rare types) were imaged at 3 T using nine b-values (0 to 800 s mm-2). Mono-exponential, stretched exponential, kurtosis, statistical, and bi-exponential models were fitted. Model preference was assessed using Bayesian Information Criterion analysis. Differences in fitted parameters between tumour types/grades and correlation between fitted parameters were assessed using two-way analysis of variance and Pearson's linear correlation coefficient, respectively.
Results
Non-mono-exponential models were preferred by 83 % of tumours with bi-exponential and stretched exponential models preferred by the largest numbers of tumours. Apparent diffusion coefficient (ADC) and diffusion coefficients from non-mono-exponential models were significantly lower in poorly differentiated tumours than well/moderately differentiated tumours. α (stretched exponential), K (kurtosis), f and D* (bi-exponential) were significantly different between tumour types. Strong correlation was observed between ADC and diffusion coefficients from other models.
Conclusions
Non-mono-exponential models were preferred to the mono-exponential model in DW-MRI data from cervical tumours. Parameters of non-mono-exponential models showed significant differences between types and grades of tumours.
Key Points• Non-mono-exponential DW-MRI models are preferred in the majority of cervical tumours.
• Poorly differentiated cervical tumours exhibit lower diffusion coefficients than well/moderately differentiated tumours.
• Non-mono-exponential model parameters α, K, f, and D* differ between tumour types.
• Micro-structural features are likely to affect parameters in non-mono-exponential models differently.