Detection of Local Recurrence in Patients with Head and Neck Squamous Cell Carcinoma Using Voxel-Based Color Maps of Initial and Final Area under the Curve Values Derived from DCE-MRI

Study Patients

A review of the data base of our institution revealed 192 consecutive patients with pathologically confirmed HNSCC and surveillance MR imaging after definitive treatment between March 2014 and May 2015. Definitive treatment included curative resection, an operation and adjuvant radiation therapy, or concurrent chemoradiation. Among these, we found 135 patients with a focal moderate to intensely contrast-enhancing lesion at the primary site on fat-suppressed contrast-enhanced (CE) T1-weighted MR images and, finally, recruited 68 patients with a final diagnosis made by pathology or clinicoradiologic follow-up of >12 months. Subsequently, an independent neuroradiologist (J.Y.L., with 5 years of experience in head and neck imaging) who was not involved in the following image interpretation and data processing checked the image quality of DCE-MR imaging and excluded 5 patients with significant metallic artifacts, leaving 63 patients as a test patient group. Three of 63 patients have been previously reported.¹⁶ This prior article dealt with histogram analysis of DCE-MR imaging to evaluate local tumor recurrence in HNSCC, whereas in the current article, we investigated the feasibility of voxel-based color maps of DCE-MR imaging. Twenty patients were randomly selected as a separate patient group for a training session to distinguish among those with an enhancing lesion at the primary site but not confirmed as either recurrence or posttreatment change, to train readers and to minimize interreader variability (Fig 1).

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

Diagram of study population enrollment.

MR Imaging Acquisition and Generation of Voxel-Based Color Maps from DCE-MR Imaging

MR imaging examinations were performed with a 3T MR imaging scanner (Magnetom Skyra; Siemens, Erlangen, Germany) using a 64-channel head and neck coil. Our MR imaging protocol for head and neck tumors consists of axial and coronal T1- and T2-weighted turbo spin-echo sequences with DCE-MR images. All of the axial T1- (TR/TE = 790/11 ms) and T2-weighted images (TR/TE = 5470/85 ms) were acquired with an FOV = 190 × 190 mm², matrix size = 448 × 291, and slice thickness = 3 mm without gap. DCE-MR imaging was performed in an axial plane with 3D controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA) technique before, during, and after administration of a standard single-bolus administration of 0.1 mmol of gadoterate meglumine (Dotarem; Guerbet, Aulnay-sous-Bois, France) per kilogram of patient body weight dose with a rate of 4 mL/s. Dynamic acquisition was performed with a temporal resolution of 3.2 seconds, and contrast was administered after 11 baseline dynamics (total = 144 dynamics). Axial (TR/TE = 650/12 ms) and coronal (TR/TE = 540/11 ms) CE-T1WIs with fat suppression were obtained after DCE-MR imaging. Detailed imaging parameters for DCE-MR imaging are as follows: slice thickness = 3 mm without gap; 20 slices; z-axis coverage = 60 mm; spatial in-plane resolution = 184 × 160; TR/TE = 6.3/3.1 ms; flip angle = 15°; FOV = 19 cm; total acquisition time = 7 minutes 24 seconds.

The DCE-MR imaging data were transferred to a personal computer to process perfusion parametric maps with an in-house plug-in program, which was developed for DCE-MR imaging data processing by calculating the trapezoidal integration of normalized signal intensity (SI)¹⁷ using ImageJ software (National Institutes of Health, Bethesda, Maryland). The contrast index was calculated for the voxels on each DCE-MR image using the following formula: Contrast Index = [SI (Postcontrast) − SI (Precontrast)] / SI (Precontrast).¹⁶ The time course of the contrast index was plotted to obtain a TSI curve and the initial and final 90-second AUC TSI values (IAUC₉₀ and FAUC₉₀) for each voxel. IAUC₉₀ was defined as the trapezoidal integration of the normalized TSI curve during the initial 90 seconds from the onset of contrast enhancement in the contrast-enhancing voxel, indicating a measurement of the initial arrival of contrast agent in the tissue of interest after intravenous bolus administration that reflects blood flow, vascular permeability, and the fraction of interstitial space.¹⁸ FAUC₉₀ was defined as the trapezoidal integration of the normalized TSI curve during the last 90 seconds in the same contrast-enhancing voxel, with IAUC₉₀, indicating the amount of contrast agent leakage within the extravascular extracellular space.¹⁹ The IAUC₉₀ and FAUC₉₀ values were then constructed to generate voxel-based color maps, which ranged from blue, green, yellow, to red (Fig 2). The SI of the enhanced tissue was displayed after normalization, relative to the unaffected prevertebral muscles on baseline images. The signal intensity of skeletal muscles was coded blue, and the vascular signal intensity exhibiting maximal enhancement was coded red. The average time for generating a pair of color maps in 1 DCE study was <2 minutes. Details for the postprocessing of the DCE-MR imaging data using the in-house-developed plug-in are described in the On-line Appendix and On-line Figs 1 and 2.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Illustration of the steps for generating color maps of initial and final 90-second AUC values using the TSI curve. A, The time course of the contrast index was plotted to obtain a TSI curve and the initial and final 90-second AUC values for each voxel. IAUC₉₀ (diagonal pattern) and FAUC₉₀ (dotted pattern) were defined as the trapezoidal integration of the normalized TSI curve during the initial and final 90 seconds from the onset of contrast enhancement in the voxel. B and C, Voxel-based color maps corresponding to IAUC₉₀ and FAUC₉₀ values were constructed in blue, green, yellow, and red.

Analysis of Conventional and DCE-MR Imaging Examinations

After reviewing axial CE-T1WI of 20 patients in the separate patient group to train readers, a neuroradiologist (K.L.C., with 7 years of experience), who was aware of the location of the primary tumor, constructed a TSI curve of DCE-MR imaging for the focal masslike enhancing lesion at the primary site using a hand-drawn ROI with ImageJ software. The ROIs were double-checked and supervised by another experienced neuroradiologist (J.H.B., with 20 years of experience). All ROIs were free from areas of necrosis or nontumor macrovessels depicted on structural images. The TSI curve was interpreted by the neuroradiologist as progressive increment (type 1), plateau (type 2), or washout (type 3). The same neuroradiologist drew a rectangular ROI on axial CE-T1WIs of the separate patient group used to train readers and the patient group to assess the diagnostic accuracy as a guide for visual assessment of voxel-based color maps by different readers.

Three neuroradiologists (J.H.L., Y.J.C., and H.W.K. with 16, 9, and 5 years of experience in head and neck imaging, respectively) conducted a visual assessment of voxel-based color maps of the separate patient group used to train readers to minimize interreader variability in their interpretation. For each patient, they assessed each pair of IAUC₉₀ and FAUC₉₀ images and then determined the patterns of the color change as progressive increment (type 1), plateau (type 2), and washout (type 3). The results were then correlated with those of the TSI curve-pattern analysis, and discrepancies were discussed to assess agreement. After the training session, they performed an independent analysis of the patient group to assess the diagnostic accuracy blinded to the final diagnosis using the same methods. They regarded type 2 (plateau) or type 3 (washout) as a recurrent tumor and type 1 (progressive increment) as a posttreatment change.^10,15,20,21

All 3 neuroradiologists performed a blinded review of conventional MR images of the patient group to test the diagnostic accuracy to detect recurrent tumor or posttreatment change, and those results were used to investigate the added value of voxel-based color maps over conventional MR imaging. Image sets included axial T2-weighted images,T1WIs, fat-suppressed CE-T1WIs, and coronal fat-suppressed CE-T1WIs. The interpretation of conventional MR imaging was performed according to the following criteria^22⇓–24: Recurrent tumors were defined as a focal mass discriminated by the surrounding tissue, with intermediate to slightly high signal intensity on T2WI, low signal on T1WI, and moderate-to-strong enhancement after gadolinium administration. Degree of enhancement was defined as “moderate” when enhancement was weaker than mucosal enhancement but greater than skeletal muscles. “Strong enhancement” was defined when the degree of enhancement was similar to or stronger than mucosal enhancement. “Posttreatment change” was defined as a fibrous scar on any focal linear or triangular enhancing lesion with very low signal intensity on T2WIs and low signal intensity on T1WIs, and “posttreatment inflammation,” as a diffuse hyperintense abnormality on T2WIs with strong contrast enhancement. Results of conventional MR imaging analysis for the detection of recurrent tumor were classified into very probable = 1, somewhat probable = 2, somewhat unlikely = 3, and unlikely = 4.

Interpretation of both conventional and DCE-MRIs was conducted with a 1-month interval from the interpretation session for conventional MR imaging alone. To assess the added value of DCE-MR imaging, we accepted the result on DCE-MR imaging with only score 2 or 3 on conventional MR imaging. Score 1 or 4 on conventional MR imaging was accepted as a final interpretation result regardless of the pattern on DCE-MR imaging.

Statistical Analysis

Pathology or clinicoradiologic follow-up for at least 12 months was the diagnostic reference standard for interpretation of conventional MR imaging and voxel-based color maps of DCE-MR imaging. To compare clinical characteristics between groups, we conducted the χ² or Fisher exact test for categoric variables. Student t and the Mann-Whitney U test were performed for the comparison of continuous variables after checking the normality with the Kolmogorov-Smirnov test. The sensitivity, specificity, positive and negative predictive values, and accuracies of the conventional MR imaging alone and the combined interpretation of conventional and DCE-MRIs were calculated for differentiation of recurrent tumor and posttreatment change, with score 1 on conventional MR imaging or score 2 on conventional MR imaging with either type 2 or 3 on DCE-MR imaging being considered recurrence. Comparison of the sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of conventional MR imaging alone and the combined interpretation of conventional MR imaging and color maps of DCE-MR imaging was performed using a generalized estimating equation model. The Cohen κ coefficient was used to measure interreader agreement of the results of DCE-MR imaging voxel-based color maps among 3 readers and agreement of the results between the TSI curve pattern and voxel-based color maps in the training session. All statistical analyses were performed using MedCalc for Windows, Version 13.0 (MedCalc Software, Mariakerke, Belgium) and SAS 9.4 software (SAS Institute, Cary, North Carolina). The significance threshold for the differences was set at a P value < .05.