Peritumoral ADC values in breast cancer: region of interest selection, associations with hyaluronan intensity, and prognostic significance

All tumor samples were obtained during the preoperative ultrasound-guided core needle biopsy (G14) 2–4 weeks before the surgery. An automated core needle gun with a 22-mm throw (Bars Magnum, Bard Biopsy Systems) was routinely placed beside the tumor, with at least a minimum of 2–3-mm distance away from the edge of the tumor in order to always include the transitional area between the normal breast tissue and the tumor margin. Any cystic or possible necrotic areas of the tumor were avoided. The samples were immediately snap-frozen in liquid nitrogen. The biopsies were not microdissected for individual cells and hence contained both pericellular stroma and breast carcinoma cells. The HA content was evaluated by an ELISA-like method, which has been described in detail previously [22, 23]. The hyaluronan content was normalized to tissue weight and dichotomized into two groups according to the median of 3 ng/mg.

MRI examinations were performed in the prone position with a 7-element phased-array coil dedicated to breast imaging (Philips Achieva 3.0-T TX, Philips N.V.). The structural breast MRI protocol consisted of five sequences: (1) T1-weighted fast field echo (TR = 4.58 ms; TE (in phase) = 2.3 ms; in-plane resolution 0.48 mm × 0.48 mm; 257 slices; slice thickness 0.7 mm; scanning time 6 min 11 s); (2) T2-weighted turbo spin echo (TR = 5000 ms; TE = 120 ms, flip angle 90°; in-plane resolution 0.6 mm × 0.6 mm; 85 slices; slice thickness 2 mm; scanning time 3 min 20 s); (3) short T1-inversion recovery/turbo spin echo (TR = 5000 ms; TE = 60 ms; TI 230 ms; in-plane resolution 1 mm × 1 mm; 90 slices; slice thickness 2 mm; scanning time 5 min 40 s); (4) a dynamic eTHRIVE sequence (TR = 4.66 ms; TE = 2.3 ms; spectrally adiabatic inversion recovery (SPAIR) fat suppression; dynamic scan time 58.5 s; in-plane resolution 0.96 mm × 0.96 mm; 180 slices; slice thickness 1 mm; with precontrast and six phases after the gadoterate meglumine (0.2 ml/kg, 3 ml/s) injection followed by a saline chaser); and (5) DWI echo planar imaging (TR = shortest; TE = 95 ms; flip angle 90°; SPAIR fat suppression; in-plane resolution 1.15 mm × 1.15 mm; 30 slices; slice thickness 4 mm; diffusion gradients in three directions; scanning time 4 min 8 s) with five respective b factors (0, 200, 400, 600, and 800 s/mm²). The ADC maps were automatically calculated linearly by the method provided by the MRI vendor.

T1-weighted, T2-weighted, and dynamic contrast–enhanced images were referred, and a crosshair tool (Sectra PACS, version 15.1.20.2, Sectra Workstation IDS7) was used to locate the lesion and to correctly position the intratumoral ROI on ADC maps. The intratumoral ROIs were placed on ADC maps with a definitive demarcation from the parenchyma and fat. The ROI was drawn polygonally to cover the entire lesion on the slice with the largest tumor diameter while cystic, necrotic, fatty, and hemorrhagic areas were carefully avoided (ROItumor).

The peritumoral ADC measurements were performed using four different ROI selection methods: method 1 (M1), a band ROI method first described by McLaughlin et al [19], with a 2-pixel width which was generated by using the ImageJ software (open-source software supported by the NIH) adjacent to the tumor border to cover the area surrounding the tumor; method 2 (M2), the whole tumor surrounding ROI method where 10-pixel-sized round ROIs were placed next to each other adjacent to the tumor border to cover the area surrounding the tumor; method 3 (M3), a simplified modification, the clockwise ROI method where eight round ROIs with 10-pixel size were placed clockwise at 0:00, 1:30, 3:00, 4:30, 6:00, 7:30, 9:00, and 10:30 positions; and method 4 (M4)., a visual assessment ROI method, first described by Mori et al [20], where three round ROIs each of 10-pixel size were placed where the ADC values visually appeared to be most increased on the breast parenchymal tissue adjacent to the tumor border. M1 was measured with ImageJ, while M2, M3, and M4 were measured directly from the picture archiving and communication systems (PACS). A schematic illustration of the four different peritumoral ROI selection methods is shown in Figs. 1c–f and 2a–d.

A breast radiologist (with 10 years of experience in breast MRI analysis) and a breast oncologist (with four years of experience in breast MRI analysis) independently measured the intratumoral and peritumoral ADC values blinded to information about histopathology and hyaluronan metabolism. Minimum, mean, and maximum values of intratumoral ADC were recorded and designated as ADCtmin, ADCtmean, and ADCtmax, respectively.

The four different methods for peritumoral area yielded various amounts of ROIs. In each peritumoral ROI, minimum, maximum, and mean ADC values were first recorded. The lowest minimum and highest maximum values in each method were selected for further analysis and designated as ADCpmin and ADCpmax, respectively. M1 yielded only one peritumoral ROI and the mean values were recorded and designated as ADCpmean. The mean values of M2–4 were averaged and also designated as ADCpmean. M4 intended to identify the maximum ADC values rather than represent the whole peritumoral area, and thus, for statistical comparison, we included only ADCpmax.

For the correlations with HA and lymph node metastasis (LNM), we could not use the highest mean from M2–4 as Mori et al [20] did because M1 gave only the mean value of the whole peritumoral area. Previous results showed that the peritumor/tumor ADC ratios could be more diagnostic than peritumoral ADC values themselves [20]. For the correlation with HA and LNM, we used ADCpmax/ADCtmean ratio.

The statistical analyses were performed using SPSS version 22 (IBM Corporation). The ADC values were evaluated as continuous dependent variables.

Intra- and interobserver reproducibility was evaluated using the interclass correlation coefficients (ICCs). An r of 1.0 was considered to indicate perfect agreement; 0.81–0.99, almost perfect agreement; 0.61–0.80, substantial agreement; 0.41–0.60, moderate agreement; 0.21–0.40, fair agreement; and ≤ 0.20, slight agreement [24].

Paired t test was performed for the comparison of tumoral ADC values between measurements with PACS and ImageJ. The Friedman test was used for comparison of peritumoral ADC values between the four ROI methods. The Mann-Whitney U test was utilized for comparisons of peritumor/tumor ADC ratios with HA status and axillary lymph node status. p values less than 0.05 were considered to be statistically significant.

Although peritumoral ADC values have been attracting more attention as a prognostic factor for breast [18‐21, 25], hepatic [26], and endometrial cancers [27], the optimal method for ROI selection is not known. The first method used in this study, the band ROI method (M1), has been previously applied in several studies [18, 19, 21, 25]. This method automatically processes almost all of the steps and does not require any specialized technique or hard-to-learn analyzing skills, which ensures straightforward access. The band ROI method also covers the whole peritumoral area in question and instantaneously gives minimum, mean, and maximum values, providing a very quick measurement. However, it does take time and effort to transfer the data from the local PACS to the separate ImageJ software, making it more difficult to apply in daily clinical practice.

In contrast, the visual assessment ROI method (M4) used by Mori et al [20] enables measurements on the local PACS software, requires no special sophisticated software function, and demands that only three ROIs have to be placed, which is less time-consuming and feasible in daily clinical practice. However, we were skeptical about the accuracy of visually selecting the highest ADC values, and suspected that the results might be operator-dependent based on the reader’s proficiency in assessing breast MR images. We therefore undertook the whole tumor surrounding ROI method (M2) in order to comprehensively measure the whole peritumoral area on local PACS, as well as to examine the clockwise 8-ROI method (M3) as its simpler modification, and compared the values obtained from those methods.

First, we confirmed that none of the minimum, mean, or maximum values of tumoral ADC showed significant differences between ImageJ and the local PACS, which indicated that these image processing programs should produce similar results. We then confirmed that the peritumoral ADC values obtained by the clockwise ROI method (M3) were equivalent to those obtained by the ROI methods surrounding the whole tumor (M2). Most importantly, the results show that only three ROIs obtained by the visual assessment ROI method (M4) were sufficient to identify the needed values. Furthermore, only the clockwise ROI method (M3) failed to provide significant results when examining the associations between peritumor/tumor ADC ratios and HA status. This can be explained by the fact that the clockwise ROI method places a limited number of peritumoral ROIs regardless of their intensities, and this caused a failure in identifying those areas with the highest representative ADC values (Figs. 1e and 2c). In contrast, the visual assessment ROI method (M4) succeeded in overcoming this obstacle and obtained the highest ADC values while requiring even fewer ROIs.

There was only a moderate ICC for peritumoral ADC measurements obtained from the clockwise method (M3). This can be explained by the fact that clockwise ROIs were difficult to place when tumor shapes were irregular (Fig. 2c). These moderate values were mostly due to the difficulty of placing ROIs on irregularly shaped tumors, where even marking the starting 12 o’clock position may prove to be a difficult task. However, the visual assessment ROI method (M4) gave almost perfect ICC, which indicates that this method is effective regardless of the reader’s experience in analyzing breast MR images. Based on our results, we conclude that visual assessment ROI measurements represent the less time-consuming and sufficiently accurate evaluation method for identifying the peritumoral area.

Although ADC has been widely investigated as an imaging prognostic biomarker, the limited reproducibility of ADC values across different imaging vendors, field strengths, and imaging centers is controversial [28‐30]. Therefore, we agree with the previous conclusions that the peritumor/tumor ADC ratio is more reliable and generally more applicable than measuring only the peritumoral area [18], and subsequently reduces bias. Furthermore, while peritumoral high intensity signals visualized on T2-weighted images were consistently shown to be associated with biologically more aggressive diseases [31, 32], this sign has only been visualized in a small portion of tumors (15–32%) [31‐33]. Accordingly, measuring peritumor/tumor ADC ratio can increase the diagnostic accuracy and efficiently evaluate the peritumoral area even in the absence of any visually obvious edema.

It was previously established that HA levels increase in most forms of inflammation, including those associated with the progression of malignant tumors. HA is a known promoter of breast cancer and other malignancies and its abundance is an indicator of an aggressive tumor type and cancer progression [12, 13]. Since HA is increased in edema and contributes to tissue hydration, we hypothesized that the HA content might be associated with the ADC ratio values. Indeed, breast cancers with a higher HA content did reveal significantly greater peritumor/tumor ADC ratios. The positive correlations of peritumor/tumor ADC ratio with lymph node positivity and HA content suggest that the ADC ratio may prove to be a valuable, non-invasive prognostic indicator available even in the preoperative setting. However, this will need to be verified in a larger patient population and with a sufficient period of follow-up.

We wanted to correlate the ADC values with tumor HA content. The assay of the HA is quite unique and relatively laborious; this meant that only a limited number of patients could be included in our analysis. Nevertheless, statistically significant results were successfully obtained; i.e., this is the first time that the association between breast tumor HA content and peritumor/tumor ADC ratios has been determined.

To conclude, the validation of the four ROI selection methods showed that all of them indicated relatively well the peritumoral ADC values of breast cancers, but a reliable result could be obtained by selecting three ROIs according to visual evaluation by the reader. The findings suggest that peritumor/tumor ADC ratios could feasibly be an applicable imaging index of the aggressiveness and prognosis of breast cancers.