Several studies have reported the usefulness of DWI in the characterization of head and neck tumors (1). DWI has been widely applied to the diagnosis and monitoring of head and neck lesions, however, their DWI (mainly ADC) and IVIM parameters varies between different intuitions and scanners (2), and their diagnostic accuracy is still limited. Therefore, there is a need to establish the gold standard in DWI and IVIM parameters to allow the robust comparison between the scanners and improve the diagnostic accuracy beyond conventional ADC. Recently, some authors have noted the importance of non-Gaussian DWI in the head and neck tumor as the additional diagnostic tool to complement ADC (3). Based on these encouraging results, our purpose was to evaluate the usefulness of non-Gaussian diffusion parameters for the clinical diagnostic ability in head and neck.

This IRB approved prospective study included 135 (62 malignant/63 benign/10 inflammation) patients suspected of head and neck tumors. Head and neck MRI was performed using a 3-T system (Skyra, Trio Tim; Siemens AG) or a 1.5-T system (Avanto; Siemens AG) equipped with a dedicated head and neck coil. The following images were obtained after localizers were acquired:

1. As the image quality of DWI for head and neck lesion suffers from severe image distortion due to susceptibility artifact, a read-out segmented EPI (RS-EPI) sequence combined with GRAPPA parallel acquisition and 2D-navigator-based reacquisition was used (4,5) with the following parameters: 9 b values of 0, 75, 150, 300, 600, 1000, 1400, 1800, 2200 sec/mm²; TR/TE 2,000/65 ms, FOV 220×220 mm², matrix 148×148, slice thickness 5.0 mm, 5 readout segments, parallel imaging factor 2, bandwidth 938 Hz/Px, echo spacing 0.36 ms and scan time 5 min 8 sec.

2. T1-weighted image: matrix size 256x256, FOV 180x180 mm², section thickness 4.0 mm; 20 sections without gap, TR/TE 700/12 ms.

3. ROIs were placed onto the lesion and images processing was performed using software implemented in Matlab (Mathwork, Natick, MA) comprising the following steps:

1/The diffusion signal acquired with b>300 s/mm² was fitted using the kurtosis diffusion model to estimate ADCo and K:

S/So = exp[-bADCo + K(bADCo)²/6] [2]

2/Then, the fitted diffusion signal component was subtracted from the corrected raw signal acquired with b<300s/mm² and the remaining signal was fitted using the IVIM model (6) to get estimates of the flowing blood fraction, fIVIM, and the pseudodiffusion, D*.

3/Synthetic ADC encompassing both Gaussian and non-Gaussian diffusion effects (2), _LbsADC_-Hb, was defined using only 2 b values as:

sADC_Lb-Hb, = ln [S(Lb)/S(Hb)]/Hb-Lb [3]

where Lb is a low “key” b value and Hb is high “key” b value are selected to provide the highest sensitivity to non-Gaussian diffusion (2). We tried 3 combinations for Lb and Hb: 0-1400, 300-1400 and 0-600 s/mm² (equivalent to a standard ADC).

The DWI and IVIM parameters in malignant and benign lesions, inflammation and LVM (Lymphatic Vascular Malformations) are shown in Table 1 and Table 2. All parameters showed the significant difference for differentiating malignant from benign lesions, and their AUC performance is shown in Figure 1. Regarding sADC and ADCo parameters, a significant difference was observed between malignant and benign lesions (p<0.01), benign lesions and inflammation (p<0.01), and malignat lesions and LVM. sADC0-1400 and sADC0-600 values in inflammation was significantly lower than in LVM (p<0.05 and p<0.01). Interestingly, K value in inflammation was significantly higher than that of malignant or benign lesions (p<0.05 and p<0.01). The combinations of low K and low ADCo values were found in eight malignant lesions, but not in inflammation. fIVIM in LVM was significantly higher than malignant or benign lesions (w/o LVM) (p<0.01 and p<0.01), Overall, the AUC performance of sADC0-1400 was highest as 0.84 with 74.2% sensitivity and 80.3% specificity.AUC of diffusion parameters for differentiating malignant and benign lesions was reasonable, considering the data derived from different scanners. Distortion artifacts were remarkably improved with RS-EPI in most cases, which enabled more approximate estimation. The combinations of both low K and ADCo values in some malignant cases suggest the different tumor characteristics other than inflammation. Higher fIVIM in LVM might explain higher perfusion fraction. Although more work is needed to conclude these results, non-Gaussian DWI and IVIM parameters have the potential to improve the diagnostic accuracy by complementing their diagnostic abilities.