Primary hyperparathyroidism is among the most common endocrine disorders with a prevalence reaching 0.1% to 0.4% [1] [2] . Diagnosis of primary hyperparathyroidism is based on biochemical markers, including serum parathyroid hormone (PTH) and serum calcium [3] [4] . With the introduction of minimally invasive unilateral surgical techniques, preoperative localization of these lesions has become part of the routine clinical workup [5] - [7] . Multiple imaging modalities have been used over the past decades for this purpose, but parathyroid scintigraphy is considered as the modality of choice according to many experts [8] [9] .

Several imaging protocols have been described for the scintigraphic evaluation of parathyroid tissue, utilizing different radiopharmaceuticals and acquisition techniques. As a rule, these protocols fall into two broad categories: those that exploit differences in the rate of washout of ^99mTc-sestamibi between thyroid tissue and parathyroid adenomas, and those that utilize 2 disparate radiopharmaceuticals to contrast functional aspects of tissue in the thyroid bed. In this latter category of dual-radiopharmaceutical (DR) imaging (often referred to as dual-iso- tope imaging), radiopharmaceuticals such as ²⁰¹Tl-thallous chloride or ^99mTc-sestamibi are used to identify areas of cellularity in the thyroid bed, while ¹²³I-NaI and ^99mTc-, which are substrates of the sodium iodide symporter, are used to identify thyroid tissue [10] [11] . Areas of cellular tissue which do not exhibit trapping of iodide-like substrates are suspect for parathyroid adenoma [12] [13] .

DR images may be acquired sequentially, if radiopharmaceuticals with disparate energies are used, simultaneously, utilizing a dual-energy acquisition technique [10] . Diagnosis is based on comparing and contrasting the distribution of the two radiopharmaceuticals, either in a simple side-by-side comparison or by way of a more sophisticated process of image registration and graded subtraction. Both the current SNMMI Practice Guidelines [10] and EANM Parathyroid Guidelines [11] advocate use of digital subtraction over simple comparison alone. Many [14] -[17] , though not all [18] , published studies assessing the diagnostic performance of DR parathyroid imaging have also employed software-based digital subtraction. Digital subtraction, when performed using images which are not intrinsically coregistered, is subject to errors caused by potential subtle variations in patient position [19] . This pitfall has motivated some physicians to rely primarily on visual comparison and not risk introduction of software-based artifacts [18] .

Interestingly, very few studies have formally compared visual and digital subtraction in DR parathyroid imaging, and of these, no advantage of digital subtraction has been demonstrated. Liehn et al. compared visual assessment to digital subtraction and were unable to demonstrate improvement in sensitivity and inter-observer reproducibility with the use of the latter technique [20] . Chen et al. also showed no increase in sensitivity and specificity with use of computer assisted digital subtraction [21] . Both of the investigations studied the effect of digital subtraction on planar images which were sequentially acquired which, as discussed, had potential for introducing errors of misregistration. In the present study, we have evaluated the effect of digital subtraction on simultaneously-acquired ¹²³I-NaI and ^99mTc-sestamibi images, a technique which yields intrinsically coregistered data which are not subject to registration errors.

90 patients met the inclusion criteria and were included in this study. The average age was 60.5 ± 12.8 years and 68.9% (62/90) of subjects were female. Average serum PTH level before imaging was 24.9 ± 22.9 pmol/L (normal 1.6 - 9.3 pmol/L). 99% (89/90) of patients had solitary parathyroid lesions and 1% (1/90) had two parathyroid adenomas, as confirmed by histopathological examination. Of these 91 lesions, 86% (78/91) were adenoma, 13% (12/91) were hyperplasia and 1% (1/91) was a parathyroid carcinoma. Mean parathyroid lesion size was 2.0 ± 0.85 cm in largest diameter based on histopathological examination. Table 1 shows the sensitivity, specificity and AUC for visual comparison and digital subtraction. All readers had significantly greater sensitivity with digital subtraction compared to visual comparison alone. Three out of 5 readers exhibited a significantly greater AUC with digital subtraction compared to visual comparison alone.

When using digital subtraction, ICC was greater among trainees (0.8399 vs. 0.7484, p = 0.0008), among expe-

^*p < 0.05.

rienced readers (0.8973 vs. 0.8492, p = 0.0062), and among all readers (0.8632 vs. 0.7648, p = 0.0001) as compared to visual interpretation alone. Finally, readers were more definitive in their interpretation of the studies with the digital subtraction of ROIs compared to visual interpretation alone (92% vs. 85%, p = 0.0365).

The diagnostic performance of DR imaging was significantly improved with the use of digital subtraction. Sensitivity of all readers significantly increased with digital subtraction compared to visual interpretation alone, without significant effect on specificity. Furthermore, the AUC was significantly greater for 3 out of 5 readers. These findings held true for both experienced readers and novice trainees. There was also significant improvement of inter-observer agreement when using digital subtraction compared to visual interpretation alone for both experienced readers and trainees. Based on the Landis et al. interpretation [24] , the inter-observer agreement improved from substantial to almost perfect in trainees.

These results are discordant from previously published data [20] [21] . We hypothesize that this is related to the fact that our study used simultaneous co-imaging thereby providing intrinsically coregistered images; this was not the case in the referenced articles, where subtle misregistration could have reduced the accuracy of the subtracted images.

A potential limitation inherent to the statistical methodology used in this study is that every ROI was analyzed individually and independently, which differs from routine clinical reading. Such an approach could minimize the satisfaction of search effect [25] , leading to improved sensitivity and potentially decreased specificity. However, given that the sample contained only one patient with 2 lesions, this effect should be minimal. Another potential limitation of this study is that the study population was confined to patients with confirmed parathyroid adenomas at surgery. This selection might prevent the inclusion of patients with equivocal imaging findings who did not proceed to parathyroidectomy. Nonetheless, there is no reason to believe that inclusion of these more equivocal cases would have favored the visual over the image subtraction method.

The sensitivities obtained in this study are relatively low, ranging from 73% to 80%, but comparable to studies using only DR planar imaging [26] . Because the purpose of this study was to evaluate the added value of digital subtraction, delayed planar images and SPECT-CT were purposely excluded, resulting in lower sensitivity. It is important to emphasize that, even with the use of digital subtraction, utilization of delayed images and/ or SPECT-CT images remains recommended [26] [27] .

The results of this study, performed using intrinsically co-registered and simultaneously ^99mTc-sestamibi and ¹²³I-NaI images, are the first to rigorously demonstrate an advantage of using digital subtraction over visual assessment alone for DR parathyroid scintigraphy. Use of subtraction imaging in this study resulted in improved inter-observer agreement, superior accuracy and greater confidence for both experienced readers and trainees.