Inclusion
In this study, we have included 69 consecutive women with primary stage II–IV ductal primary breast cancer who had been referred for a pre-treatment whole-body FDG-PET/CT scan. We only included patients with at least one loco-regional lymph node which could be classified as benign or malignant based on histological or imaging information. We received a waiver from the Medical Ethical Committee of our institution (METC Isala, Zwolle) to perform this retrospective study, as it deals with an evaluation of clinically indicated scans. Informed consent was obtained from all individual participants included in the study.
PET/CT data acquisition
Patients fasted for at least 6 h prior to scanning. Before intravenous injection of FDG, blood glucose levels were measured to ensure a value below 10 mmol/L. The mean glucose level was 5.4 mmol/L (range 3.8–9.3 mmol/L). A dedicated dose protocol depending quadratically on patients’ body weight was used. This protocol is described by the formula A = 3.8 × w2/t, where A is the FDG activity to administer (in megabecquerel), w is the patients’ body weight (in kilogram), and t is the acquisition time per bed position (in seconds). This approach has been shown to result in an image quality that does not depend on patient’s weight [13]. Acquisition times for the patient studies were 1 and 2 min per bed position for patients with body weight ≤ 80 and > 80 kg, respectively. The average administered FDG activity was 331 MBq (range 155–533 MBq).
All scans were acquired with patients in supine position, using a state-of-the-art PET/CT scanner (Ingenuity TF, Philips Healthcare, Cleveland, OH, USA). This fully three-dimensional TOF PET scanner is combined with a 128-slice CT scanner. The PET scan was acquired 60 min post-injection, using a whole-body protocol. Before PET imaging, a CT scan was acquired for attenuation correction. The CT scan parameters were tube voltage 120 kV, dose modulation with an average tube current of 53 mA (range 37–94 mA), slice collimation 64 × 0.625 mm, pitch 0.83 and rotation time 0.5 s.
PET/CT data reconstruction
PET data were reconstructed using a list-mode TOF algorithm and line-of-response row-action maximum-likelihood algorithm method [14, 15], called BLOB-OS-TF. Images were reconstructed in two types of matrices: 144 × 144 matrices with voxel size 4 × 4 × 4 mm3 (standard-voxels) and 288 × 288 matrices with voxel size 2 × 2 × 2 mm3 (small-voxels). For the standard-voxel reconstruction, the blob had a 2.5-mm radius with a blob shape parameter of 8.4 mm. The blob radius and shape parameter for the small-voxel reconstruction were 2.8 and 6.4 mm, respectively. Furthermore, the relaxation parameters for the standard- and small-voxel reconstructions were 1.0 and 0.5, respectively. For both types of voxel reconstructions, 3 iterations and 43 subsets were applied. All reconstruction parameters were default settings recommended by the vendor. Point-spread function modelling was not applied.
CT data were reconstructed using an iterative reconstruction algorithm (iDose, Philips Healthcare, Cleveland, OH, USA) with iDose level 4 and a slice thickness of 3 mm. The administered FDG activity and PET/CT acquisition protocols were consistent with European Association of Nuclear Medicine (EANM) guidelines for tumour PET imaging [16, 17]. Moreover, the reconstructed PET images with standard-voxels fulfilled the EANM research Ltd. (EARL) accreditation specifications [18]. The small-voxel reconstruction does not fulfil the EARL accreditation specifications, because the recovery curves for the small 10- to 13-mm spheres increase up to values above the maximum EARL specifications [10].
Visual evaluation
Integrated PET/CT data were reviewed on a dedicated workstation (IntelliSpace Portal 6, Philips Healthcare, Cleveland, OH, USA). First, each PET/CT scan was evaluated by two nuclear medicine (NM) physicians, with more than 5 years of experience in PET/CT viewing. They were blinded to the patient record and histological information and interpreted the PET/CT data by simultaneous viewing of PET, CT and fused PET/CT images. Both the standard- and small-voxel images were evaluated blindly, both separately and randomly. The NM physicians scored all loco-regional lymph nodes showing focal FDG-uptake on the standard- or small-voxel images. They integrated their PET reading with the presence, absence, shape and size of lymph nodes on the low-dose CT scan, in an identical fashion as used in clinical interpretation.
Initially, each lymph node was scored using a five-point ordinal scale with 1: certainly benign, 2: probably benign, 3: equivocal, 4: probably malignant and 5: certainly malignant. If this initial interpretation between both physicians differed, consensus was reached. This was needed for 39 lymph nodes (19%) on standard-voxel images and for 43 lymph nodes (21%) on small-voxels images. Next, to be able to evaluate the lymph node characterization performance, each lesion was assigned as benign or malignant using the following method. All lymph nodes with a score of 1 or 2 were allocated as benign. All lymph nodes with scores of 4 and 5 were allocated as malignant. Lymph nodes with a score of 3 were once again evaluated on the PET/CT images, and they received an ultimate score as benign or malignant.
Quantitative evaluation
All scored lymph nodes were evaluated semi-quantitatively by an experienced PET reader blinded to the patient record, histological information and visual PET/CT scores. The maximum standardized uptake value (SUVmax) was derived on the axial slice that contained the highest FDG-uptake of the lesion.
Next, we calculated the lymph node-to-background ratio (TBratio), defined as the ratio between the lymph node SUVmax and the average SUV in the background (SUVbackground). To measure the SUVbackground, we defined two regions of interest (ROI1 and ROI2) on the axial PET image. ROI1 enclosed both the lymph node under study and a surrounding background area of 800 mm2, while ROI2 only enclosed the lymph node. For both ROIs, the area size and the average SUV (SUVmean) were collected to calculate the SUVbackground in a donut-shaped ROI using Formula 1:
$$ {\mathrm{SUV}}_{\mathrm{background}}=\frac{\left(\mathrm{ROI}1\ {\mathrm{SUV}}_{\mathrm{mean}}\bullet \mathrm{ROI}1\ \mathrm{area}\right)-\left(\mathrm{ROI}2\ {\mathrm{SUV}}_{\mathrm{mean}}\bullet \mathrm{ROI}2\ \mathrm{area}\right)\ }{\mathrm{ROI}1\ \mathrm{area}-\mathrm{ROI}2\ \mathrm{area}} $$
(1)
Finally, for all the scored lymph nodes, we measured the short-axis diameter on the axial slice of the attenuation CT scan.
Final diagnosis
The final diagnosis for each lymph node was based on histological information, follow-up (FU) imaging (FDG-PET/CT, contrast-enhanced CT or magnetic resonance imaging (MRI)) or additional imaging (contrast-enhanced CT or MRI) in the following way (Fig. 1). For patients who initially underwent a surgical resection that included sentinel lymph node biopsy or axillary lymph node dissection, the final diagnosis was based on the histological information obtained during surgery. Pathology examination was part of the clinical evaluation and was centralised at our institution. Lymph nodes were histologically processed by formalin fixation followed by paraffin embedding, according to standardized procedures.
The lymph nodes were serially sectioned at 250 μm at three levels and stained with both hematoxylin and eosin, with an immune-histochemical cytokeratin staining (panCK). The immune-histochemical procedure was performed by a fully automated procedure, using pre-diluted antibodies on the Ventana Benchmark system (Roche Ventana, Tucson AZ, USA). The sizes of the metastases were measured on a conventional bright-field microscope (Leica, DM4000, Leica microsystems Germany) using a micro-measuring scale on glass slide (definition = 0.1 mm). In all lymph nodes, the largest diameter of a metastasis was reported.
For patients who were treated with neo-adjuvant chemotherapy, the final diagnosis was based on the response to therapy as visualised on FU imaging combined with the histological information that was available from the subsequent surgical resection. For these patients, lesions were considered malignant when they showed a decrease in size or FDG-uptake induced by subsequent chemotherapy. Furthermore, lesions that were stable in size and FDG-uptake during neo-adjuvant therapy were considered to be benign unless there was proof of malignancy from histological information obtained during surgery. When histological information or FU imaging was not available, the final diagnosis was based on the results of additional contrast-enhanced CT or MRI.
Additionally, we collected information on all malignant loco-regional lymph nodes that were found during surgical resection but which had not been visualised on FDG-PET or the attenuation CT. For those lesions, we recorded the metastatic deposit size that was measured during a separate pathology examination, performed by one pathologist (JB). For visual PET performance evaluation, these lymph nodes were regarded as benign nodes on PET. Furthermore, for quantitative PET evaluation, these lymph nodes were not taken into account because it was not possible to perform measurements on PET images.
Lymph node characterization
Visual evaluation scores were analysed on a lesion-per-lesion basis, by comparing the scores on standard- and small-voxel images for each lymph node. Quantitatively, we calculated average values for SUVmax and TBratio in both benign and malignant lymph nodes and for both voxel reconstructions. We created receiver operator curves (ROC) and calculated the area under the curve (AUC) with a 95% confidence interval (CI) for SUVmax and TBratio. For both reconstruction methods, we determined the sensitivity, specificity and accuracy for lymph node characterization from the visual and quantitative PET/CT evaluation, using the final diagnosis as a reference standard. We calculated optimal cut-off values for SUVmax and TBratio to distinguish benign from malignant lymph nodes on both voxel reconstructions. These cut-off values were based on the highest combined sensitivity and specificity (highest sum).
Statistical analysis
We used the McNemar test for paired samples to compare the visual scores for both reconstructions with the final diagnosis. Quantitative results were presented as mean ± standard deviation (SD). We included ranges in uptake values and lymph node size. Differences in SUVmax and TBratio between benign and malignant lymph nodes were evaluated using the Mann-Whitney U test. Furthermore, to evaluate differences in characterization performances between standard and small-voxels for SUVmax and TBratio, we compared the AUCs using a chi-square test. Additionally, the characterization performances for SUVmax and TBratio using optimal cut-off values were evaluated with the McNemar test for paired samples. A p value less than 0.05 was considered to indicate statistical significance.
