Comparison of Diagnostic Performance Between Standard Susceptibility Map‐Weighted Imaging and Susceptibility Map‐Weighted Imaging Reconstructed From Clinical SWI and Neuromelanin MRI in Early Parkinson's Disease

ABSTRACT

Background

Susceptibility map‐weighted imaging (SMwI) provides high accuracy for early Parkinson's disease (ePD) by enhancing nigral hyperintensity. However, standard SMwI requires dedicated acquisition; reconstruction from routinely acquired sequences may offer a practical alternative.

Purpose

To compare the diagnostic performance of standard SMwI with SWI‐ and neuromelanin (NM)‐driven SMwI for differentiating ePD from disease controls (DC).

Study Type

Retrospective.

Subjects

One hundred and eighty‐seven drug‐naïve ePD (age: 67.5 ± 9.4 years, 102 male) and 43 DC (age: 67.1 ± 9.2 years, 12 male).

Field Strength/Sequence

3 T, multi‐echo gradient‐echo (GRE) for standard SMwI; GRE‐based NM imaging: multi‐echo GRE for SWI.

Assessment

Nigral hyperintensity was assessed by three neuroradiologists (4‐, 2‐, and 1‐year experience). Diagnostic performance was evaluated using clinical diagnosis for all subjects and PET for 106 subjects (98 ePD, 8 DC). PET‐based assessment included side‐based analyses accounting for PET laterality and patient‐based analyses regardless of direction.

Statistical Tests

Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy and area under the curve (AUC) were assessed using clinical diagnosis as reference; sensitivity, specificity, and accuracy using PET. Significance was set at p < 0.05.

Results

Using clinical diagnosis as reference, standard SMwI showed highest diagnostic performance (sensitivity 0.856; specificity 1.000; PPV 1.000; AUC 0.928). SWI‐driven SMwI maintained similar sensitivity (p = 0.355) but showed significantly reduced specificity, PPV, accuracy, AUC. In the PET subgroup, side‐ and patient‐based analyses showed no differences in sensitivity (p = 0.261, p = 0.670), specificity (p = 1.000, p = 1.000) and accuracy (p = 0.268, p = 0.695) between standard and SWI‐driven SMwI.

Data Conclusion

SWI‐driven SMwI approached the diagnostic performance of standard SMwI with PET reference but showed inferior performance with clinical diagnosis, suggesting a fallback option when dedicated SMwI is unavailable.

Evidence Level

3.

Technical Efficacy

Stage 1.

Plain Language Summary

Early Parkinson's disease is often difficult to diagnose. Susceptibility map‐weighted imaging (SMwI) is a useful MRI technique, but the standard method requires a dedicated scan that is not always available in routine practice. In this study, we compared standard SMwI with SMwI reconstructed from commonly acquired MRI sequences. SMwI reconstructed from susceptibility‐weighted imaging showed diagnostic performance similar to standard SMwI when PET imaging was used as the reference, but inferior performance when clinical diagnosis was used due to lower specificity. This method may be a practical alternative when standard SMwI cannot be performed, while neuromelanin‐based SMwI showed lower performance.

1. Introduction

Parkinson's disease (PD) is a progressive neurodegenerative disorder whose prevalence is rising worldwide, creating a substantial socioeconomic burden [1, 2]. Early and accurate diagnosis is important for timely intervention, yet clinical assessment alone remains challenging in early‐stage disease, where motor symptoms may be both subtle and non‐specific [3, 4]. In particular, several related disorders can mimic parkinsonism even without definite dopaminergic degeneration [5, 6]. Dopaminergic imaging such as ¹⁸F‐N‐3‐fluoropropyl‐2‐β‐carbomethoxy‐3‐β‐(4‐iodophenyl)nortropane (¹⁸F‐FP‐CIT) PET improves diagnostic confidence but is limited by accessibility, cost, and suitability for screening [7].

Thus, the continuous demand for reliable biomarkers for PD has persisted, and nigrosome imaging is an emerging biomarker for PD. Nigrosomes represent calbindin‐poor subregions within the substantia nigra pars compacta containing densely packed dopaminergic neurons and are classically divided into five nigrosomes [8]. Nigrosome 1 is the largest of the five. It is located in the dorsolateral aspect of the substantia nigra, parallel to the rostrocaudal axis of the substantia nigra. Histopathologic studies have revealed that approximately 98% of dopaminergic neurons are lost in nigrosome‐1 of symptomatic PD [9]. On susceptibility‐based MRI, a dorsolateral nigral hyperintensity (“swallow‐tail sign”) is commonly used as an imaging marker of nigrostriatal degeneration [10, 11]. Furthermore, in PD, loss of nigral hyperintensity has been considered a potential biomarker [12].

Susceptibility map‐weighted imaging (SMwI) is a promising MRI technique for PD by improving the conspicuity of nigral hyperintensity with application of quantitative susceptibility mapping (QSM)‐derived weighted factors to magnitude images [10, 13, 14]. Prior studies have reported high diagnostic accuracy of SMwI, comparable to dopaminergic PET imaging, suggesting its potential as a non‐invasive and widely available alternative [12, 15, 16]. However, standard SMwI typically requires a dedicated multi‐echo gradient‐echo acquisition in an oblique‐coronal orientation with thin slices in order to minimize partial volume effects [14]. Also, in order to improve SNR and CNR of phase images, long echo time to obtain sufficient phase contrast is required [13]. These technical requirements lengthen acquisition time of the source image and may hinder implementation.

A practical solution is to reconstruct SMwI from sequences that are already acquired for other purposes. Susceptibility‐weighted imaging (SWI) is widely incorporated into routine brain MRI protocols for the evaluation of hemorrhage, iron deposition, and venous structures, and thus represents an attractive candidate for reconstruction [17, 18]. Neuromelanin (NM) imaging, in contrast, is not yet routine and is generally considered to be an advanced sequence. Nonetheless, NM imaging is increasingly adopted in PD evaluations because it directly visualizes signal loss in the neuromelanin‐rich substantia nigra pars compacta [19, 20, 21, 22, 23]. If SMwI could be reconstructed from NM imaging, clinicians would gain access to two complementary biomarkers—neuromelanin loss and nigral hyperintensity—without additional scan time, potentially improving diagnostic confidence. The hypothesis of this study was that while dedicated standard SMwI would provide superior diagnostic performance to reconstructed SMwI, SWI‐ or NM‐driven SMwI could still offer clinically acceptable alternatives, particularly when dedicated SMwI cannot be acquired in routine practice.

Thus, the aim of this study was to compare the diagnostic performance of standard SMwI with SMwI reconstructed from SWI and NM imaging for differentiating early PD from disease controls. A further aim was to perform side‐based and patient‐based analysis in a subgroup of patients with ¹⁸F‐FP‐CIT PET with a view to determining whether SWI‐ or NM‐driven SMwI could offer a clinically acceptable alternative.

2. Materials and Methods

2.1. Subjects and Clinical Assessment

This retrospective study was approved by the Institutional Review Board of Samsung Medical Center (IRB No. SMC2025‐10‐011), and the requirement for informed consent was waived by the board. Consecutive patients who visited the movement disorders clinic at Samsung Medical Center with suspected parkinsonism between October 2021 and December 2022 were screened. All participants satisfied the following criteria: (1) Age > 18 years; (2) Diagnosis of PD based on the clinical diagnostic criteria of Movement Disorder Society (MDS) Clinical Diagnostic Criteria for Parkinson's disease [7], and parkinsonian symptoms were evaluated with the Unified Parkinson's Disease Rating Scale‐III (UPDRS‐III) [24], and the modified Hoehn and Yahr (H&Y) scale established by a multidisciplinary panel consisting of movement disorder neurologists and by consensus review of clinical history, neurological examination, and longitudinal follow‐up data; (3) Early idiopathic Parkinson's disease (PD) defined as disease duration < 5 years and Hoehn & Yahr (H&Y) stage < 3; (4) Drug‐naïve at the time of MRI acquisition, confirmed from clinical records and prescription data; (5) Absence of other concomitant neurological or psychiatric disorders; (6) No history of intracranial surgery or traumatic brain injury. Disease control (DC) patients who underwent the same brain MRI protocol of the same period were included. Patients diagnosed with essential tremor, vascular parkinsonism, cerebellar ataxia, or physiologic tremor were included. Exclusion criteria were as follows: (1) Participants with a history of major neurological, psychiatric, or systemic diseases; (2) absence of any of the three sequences (SMwI, NM, or SWI).

2.2. Subject Information

A total of 241 subjects were included in this study, of whom 11 were subsequently excluded owing to missing imaging data. Of the remaining 230 subjects, 187 subjects (85 females; 102 males) were enrolled in the early PD group, and 43 subjects (31 females; 12 males) in the DC group (Table 1). The mean ages were 67.5 ± 9.4 and 67.1 ± 9.2 years for early PD and DC groups, respectively, with no significant difference (p = 0.792). For the early PD group, disease duration was 25.2 ± 16.6 months, UPDRS‐III was 16.9 ± 8.4, H&Y scale was 1.6 ± 0.5 and MDS‐UPDRS III was 16.1 ± 8.0. For the diagnostic performance evaluation comparing with ¹⁸F‐FP‐CIT PET results, 106 subjects (98 ePD and 8 DC) were included in the analysis (Table S1).

TABLE 1.

Demographic and clinical characteristics of Parkinson's disease (PD) and disease control (DC) subjects.

	PD (n = 187)	DC (n = 43)	p
Age (years)	67.5 ± 9.4	67.1 ± 9.2	0.792
Sex (female:male)	85:102	31:12	0.002*
Disease duration (months)	25.2 ± 16.6
UPDRS‐III	16.9 ± 8.4
H&Y scale	1.6 ± 0.5
MDS‐UPDRS III	16.1 ± 8.0

Note: Student's t‐test was applied for the comparison of age and Chi‐square test were utilized for the comparison of sex. Data are mean ± standard deviation or number of patients. p‐values less than 0.05 are regarded significant (*).

Abbreviations: H&Y scale, Hoehn and Yahr scale; UPDRS‐III, modified Unified Parkinson's Disease Rating Scale‐III.

2.3. Standard SMwI Protocol and Post Processing

Nigral imaging was performed using a 3D multiecho spoiled gradient‐echo sequence on a 3 T MRI scanner (Ingenia CX, Philips Healthcare, Best, The Netherlands). Images were acquired in an oblique‐coronal plane, perpendicular to the midbrain structure. The sequence parameters were: repetition time (TR) = 49 ms, monopolar multi‐echo GRE readout, number of echoes = 3, first echo time (TE) = 27 ms, echo spacing = 13 ms, pixel bandwidth = 100 Hz/pixel, acquired spatial resolution = 0.5 × 0.5 mm², field of view = 192 × 192 mm², slice number = 32, slice thickness = 1 mm, slice gap = 0 mm, and flip angle = 20°, approximate scan time = 293 s. The postprocessing to reconstruct the standard SMwI was performed using MatLab (MathWorks, Natick, MA, USA). The QSM was calculated with STI Suite software [25, 26]. All source images underwent 3D phase unwrapping, background phase removal and dipole deconvolution to calculate magnetic susceptibility. 3D phase unwrapping and HARmonic (background) PhasE Removal using the Laplacian operator and SHARP with varying spherical kernel sizes method were implemented for phase processing [25, 26]. Improved sparse linear equation and least squares (LSQR) algorithm were then applied for QSM calculation. In the QSM calculation pipeline, the same pipeline was applied across the QSM reconstruction process using three different sequences and there was no additional parameter tuning or optimization performed outside the software framework. The calculated QSM was re‐processed to produce a QSM mask, required for the susceptibility contrast weighting [14]. Multiecho combined magnitude images were then combined with these masks to generate the standard SMwI.

2.4. SWI and NM Protocols and Post Processing

SWI and NM imaging were acquired on the same scanner with the default clinical protocols implemented at our institution which was intentional as the aim of this study was to reconstruct SMwI from imaging routinely available in standard clinical workflow. SWI was acquired in the axial anterior commissure/posterior commissure (AC/PC) line using a multi‐echo 3D gradient‐echo sequence (sequence parameters: TR = 31 ms, monopolar multi‐echo GRE readout, number of echoes = 4, first TE = 7.2 ms, echo spacing = 6.2 ms, pixel bandwidth = 255 Hz/pixel, spatial resolution = 0.5 × 0.5 mm², field of view = 240 × 240 mm, slice number = 160, slice thickness = 2 mm, gap = 0 mm, and flip angle = 17°, approximate scan time = 258 s). NM images were obtained in the imaging slab oriented perpendicular to the 4th ventricle, centered tangential to the top of the pons, using a simple 3D gradient‐echo‐based neuromelanin sequence [27]. The sequence parameters were as follows: TR = 60 ms, monopolar multi‐echo GRE readout, number of echoes = 5, first TE = 3.9 ms, echo spacing = 5.9 ms, pixel bandwidth = 178 Hz/pixel, spatial resolution = 0.7 × 0.7 mm², field of view = 256 × 256 mm², slice number = 64, slice thickness = 1.2 mm, slice gap = 0 mm, and flip angle = 20°, approximate scan time = 310 s. Using the same pipeline as above, SWI‐driven SMwI and NM‐driven SMwI were reconstructed from the respective source images. All sequences were acquired within a single imaging session. The acquisition order in our routine protocol was as follows: 3D T1‐weighted imaging, NM imaging, dedicated standard SMwI, SWI, diffusion‐weighted imaging, axial T2‐weighted imaging, and axial FLAIR imaging.

2.5. ¹⁸F‐FP‐CIT PET Acquisition

Among 230 subjects enrolled in this study, 106 subjects (98 ePD and 8 DC subjects) underwent ¹⁸F‐FP‐CIT PET. Brain ¹⁸F‐FP‐CIT PET images were obtained using a PET/computed tomography (CT) scanner (Discovery STE, GE Healthcare, Milwaukee, Wisconsin, USA), 120 min after the injection of 185 MBq ¹⁸F‐FP‐CIT. PET images were acquired for 10 min in three‐dimensional mode immediately after a brain CT scan for attenuation correction and image fusion. CT scanning was performed at 140 kVp and 50 mAs, with a slice thickness of 3.75 mm. The ¹⁸F‐FP‐CIT PET images were reconstructed by a 3D ordered‐subsets expectation maximization (OSEM) algorithm (20 subsets, 2 iterations) using a 128 × 128 matrix. All ¹⁸F‐FP‐CIT PET images were re‐evaluated in formal research read formats by an experienced nuclear medicine physician (S.H.M.; 15 years of experience), who was blinded to the MRI findings. Each scan was classified as normal or abnormal for each side of striatum.

2.6. Image Analysis

Three neuroradiologists (B.S., Y.S., and S.L.; 4, 2, and 1 year of SMwI experience) independently and blindly assessed nigral hyperintensity on both substantia nigra (SN) for standard SMwI, SWI‐driven SMwI, and NM‐driven SMwI. Prior to the formal image review, an independent training dataset consisting of 20 subjects (10 patients with Parkinson's disease and 10 healthy controls) was used to familiarize the three neuroradiologists with the imaging features of nigral hyperintensity. This training set was not included in the main study cohort or statistical analyses. Nigral hyperintensity was defined as a linear or ovoid hyperintense focus on the dorsolateral SN surrounded by hypointensity (Figure 1) [12, 14, 16, 28, 29]. Clinical diagnoses and PET results were unavailable to raters at the time of image review. To minimize selection‐ or experience‐related bias, three neuroradiologists reviewed the images and majority‐vote rating was done for the final analyses. Side‐based analysis was defined as a side‐based evaluation, in which the right and left striatum were assessed separately for the presence of dopaminergic deficit. Patient‐based analysis was defined as overall classification based on whether either side demonstrated abnormal ¹⁸F‐FP‐CIT PET uptake.

FIGURE 1.

Representative examples of nigral hyperintensity in (a) an 81‐year‐old female patient with early Parkinson's disease and (b) a 62‐year‐old female disease control subject across three reconstructions: Standard SMwI, SWI‐driven SMwI, and NM‐driven SMwI. Arrows indicate the dorsolateral substantia nigra. In the disease control subject, preserved dorsolateral nigral hyperintensity is clearly visualized on the standard SMwI, whereas this hyperintensity appears less conspicuous with blurred margins on the SWI‐driven SMwI and shows reduced overall susceptibility contrast on the NM‐driven SMwI. In contrast, the patient with early Parkinson's disease demonstrates loss of dorsolateral nigral hyperintensity (arrowheads) on the standard SMwI, with less distinct depiction on the reconstructed images.

2.7. Statistical Analysis

All analyses were performed using the R Statistical Software (v4.1.2; R Core Team 2021). For the demographic and clinical evaluation, Student's t‐tests and Chi‐square tests were applied for the comparison of early PD and DC groups, and Kruskal–Wallis tests and Chi‐square tests were utilized for the comparison of demographic data of ¹⁸F‐FP‐CIT PET subgroup. Inter‐rater agreement was assessed using Fleiss' kappa coefficient and the reference was set as very good (0.75–1.00), fair to good (0.41–0.75) and poor (less than 0.40). For the primary analysis, the results of the most experienced observer (4 years of experience in SMwI) were used, while the assessments from the other two observers were used to evaluate inter‐rater variability. The diagnostic performances of the SMwI, SWI‐driven SMwI and NM‐driven SMwI were assessed based on clinical diagnoses. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy and area under the receiver operating characteristic curve (AUC) were calculated, and the results of SWI‐ and NM‐driven SMwIs were compared with that of the standard SMwI. For the comparison of sensitivity, specificity, and accuracy, McNemar's test was performed; for PPV and NPV, generalized estimating equations were used; and for AUC, Delong's method was applied. Additional subgroup analysis was performed for the patients with ¹⁸F‐FP‐CIT PET results. Side‐based analysis was performed using logistic regression analysis with generalized estimating equations to account for within‐subject correlation between sides, and patient‐based paired comparisons were performed using McNemar's test and Delong's method. All tests were two‐sided and statistical significance was set at p‐value < 0.05.

3. Results

3.1. Inter‐Rater Agreement on Assessment of SMwI

Fleiss' κ values indicated fair to good agreement across three raters for standard SMwI (κ = 0.70), SWI‐driven SMwI (κ = 0.61), and NM‐driven SMwI (κ = 0.51).

3.2. Diagnostic Performance of SMwI Versus Clinical Diagnosis

When the clinical diagnosis was used as the reference standard, the standard SMwI achieved the highest diagnostic performance (Figure 2 and Table 2). The sensitivity, specificity, PPV, NPV, accuracy and AUC were 0.856, 1.000, 1.000, 0.614, 0.883, and 0.928. In contrast, the NM‐driven SMwI demonstrated inferior results, with reduced sensitivity, NPV, and accuracy, AUC of 0.668, 0.404, and 0.726, 0.823 with a statistically significant difference and specificity, and PPV of 0.977 (p = 0.317), and 0.992 (p = 0.307) without a statistically significant difference. The SWI‐driven SMwI maintained relatively high sensitivity and NPV of 0.824 (p = 0.355) and 0.515 (p = 0.076) with statistically significantly reduced specificity, PPV, accuracy, and AUC of 0.814, 0.950, 0.822, and 0.819. The detailed performances of other raters are demonstrated in Tables S2 and S3.

FIGURE 2.

Summary of diagnostic performance in reference to clinical diagnosis for all raters (R1, R2, and R3). Points indicate AUCs with 95% CIs for standard, SWI‐driven, and NM‐driven SMwIs of three raters; p‐values denote DeLong comparisons versus the standard SMwI and are marked with * when statistically significant. NM‐driven SMwI approached standard SMwI with no statistically significant difference observed for raters 1 and 3. SWI‐driven SMwI showed persistently lower AUC for all 3 raters. Differences in overall diagnostic performance should be interpreted in conjunction with sensitivity and specificity profiles.

TABLE 2.

Diagnostic performance of standard, NM‐driven and SWI‐driven SMwIs using clinical assessment as reference standard for rater 2.

	Standard	NM‐driven	SWI‐driven	p
				NM vs. standard	SWI vs. standard
Sensitivity	0.856 (0.798, 0.899)	0.668 (0.598, 0.732)	0.824 (0.763, 0.871)	< 0.0001*	0.355
Specificity	1.000 (0.918, 1.000)	0.977 (0.879, 0.996)	0.814 (0.674, 0.903)	0.317	0.005*
PPV	1.000 (1.000, 1.000)	0.992 (0.977, 1.000)	0.950 (0.918, 0.984)	0.307	0.005*
NPV	0.614 (0.496, 0.720)	0.404 (0.314, 0.501)	0.515 (0.397, 0.631)	< 0.0001*	0.076
Accuracy	0.883 (0.835, 0.918)	0.726 (0.665, 0.780)	0.822 (0.767, 0.866)	< 0.0001*	0.048*
AUC	0.928 (0.903, 0.953)	0.823 (0.782, 0.863)	0.819 (0.754, 0.884)	< 0.0001*	0.002*

Note: Data are presented with a 95% confidence interval. p‐values less than 0.05 are regarded significant (*).

Abbreviations: AUC, area under the receiver operating characteristic curve; NM, neuromelanin imaging; NPV, negative predictive value; PPV, positive predictive value; SMwI, susceptibility map‐weighted imaging; SWI, susceptibility weighted imaging.

3.3. Comparison of Diagnostic Performance Using Standard, SWI‐Driven, and NM‐Driven SMwIs

The standard SMwI outperformed NM‐driven SMwI, showing significantly higher sensitivity (0.856 vs. 0.668), NPV (0.614 vs. 0.404), and accuracy (0.883 vs. 0.726), AUC (0.928 vs. 0.823). In comparison with SWI‐driven SMwI, standard SMwI achieved superior specificity (1.000 vs. 0.814), PPV (1.000 vs. 0.950), accuracy (0.883 vs. 0.822), and AUC (0.928 vs. 0.819).

3.4. Subgroup Analysis With ¹⁸F‐FP‐CIT PET

When ¹⁸F‐FP‐CIT PET was used as the reference, for rater 2, side‐based analysis showed that standard SMwI yielded significantly higher sensitivity (0.787 vs. 0.617) but lower specificity without statistical significance (0.828 vs. 0.966; p = 0.058) than NM‐driven SMwI, whereas no significant differences (sensitivity p = 0.261; specificity p = 1.000; accuracy p = 0.268) were observed between standard and SWI‐driven SMwI (Table 3). In patient‐based analysis, standard SMwI again outperformed NM‐driven SMwI, with significantly higher sensitivity (0.682 vs. 0.506), whereas specificity (0.762 vs. 0.952) was significantly lower in standard SMwI. There were no significant differences between standard and SWI‐driven SMwI in patient‐based analyses (sensitivity p = 0.670; specificity p = 1.000; accuracy p = 0.695; Table 3). The detailed performances of other raters are demonstrated in Tables S4 and S5.

TABLE 3.

Diagnostic performance of standard, NM‐driven and SWI‐driven SMwI methods in the ¹⁸F‐FP‐CIT PET subgroup for rater 2.

Analysis level	Metric	Standard	NM‐driven	SWI‐driven	p (std vs. NM)	p (std vs. SWI)
Patient‐based	Sensitivity	0.682 (0.577, 0.772)	0.506 (0.402, 0.610)	0.659 (0.553, 0.751)	0.001*	0.670
	Specificity	0.762 (0.549, 0.894)	0.952 (0.773, 0.992)	0.762 (0.549, 0.894)	0.046*	1.000
	Accuracy	0.698 (0.605, 0.777)	0.594 (0.499, 0.683)	0.679 (0.585, 0.760)	0.022*	0.695
Side‐based	Sensitivity	0.787 (0.710, 0.848)	0.617 (0.531, 0.697)	0.738 (0.653, 0.808)	< 0.0001*	0.261
	Specificity	0.828 (0.634, 0.930)	0.966 (0.789, 0.995)	0.828 (0.634, 0.930)	0.058	1.000
	Accuracy	0.792 (0.725, 0.847)	0.665 (0.588, 0.734)	0.750 (0.676, 0.812)	0.000*	0.268

Note: A total of 106 subjects (98 ePD and 8 DC) were included in the subgroup analysis. Data are presented with a 95% confidence interval. p‐values less than 0.05 are regarded significant (*).

Abbreviations: NM, neuromelanin imaging; SMwI, susceptibility map‐weighted imaging; SWI, susceptibility weighted imaging.

4. Discussion

In this study, standard SMwI demonstrated superior diagnostic performance compared with SWI‐driven and NM‐driven SMwI when clinical diagnosis was used as the reference standard for differentiating early idiopathic PD from DC. On the other hand, when ¹⁸F‐FP‐CIT PET was used as an imaging reference, the performance of SWI‐driven SMwI approached that of the standard protocol, with no statistically significant differences observed in sensitivity, specificity, or accuracy, whereas NM‐driven SMwI generally showed inferior performance. Overall, these findings indicate that reconstructed SMwI, although attractive from a workflow perspective, does not yet consistently match the diagnostic reliability of dedicated SMwI across clinical reference standards. Importantly, this study exclusively included drug‐naive patients with early PD, minimizing potential confounding effects of dopaminergic medication on both MRI contrast and PET findings.

These results align with prior studies reporting high accuracy of SMwI for early PD diagnosis and its comparability to dopaminergic imaging in selected settings [30, 31, 32, 33]. For example, Sung et al. [30] reported no significant difference between SMwI and ¹⁸F‐FP‐CIT PET on both patient‐based and side‐based analyses. Bae et al. [33] compared SMwI and SWI based diagnoses of PD and showed that SMwI provided significantly superior diagnostic performance (AUC 0.791 vs. 0.720). Wang et al. [32] reported that SMwI based diagnosis of early PD achieved a sensitivity and specificity of 94.59% and 92.49%. While these prior studies established the overall diagnostic value of SMwI and its comparability to dopaminergic imaging, the current study specifically addressed whether SMwI reconstructed from clinically available SWI or NM sequences could serve as a substitute for the dedicated higher‐resolution protocol.

Two technical factors, acquisition geometry and spatial resolution, possibly underlie the observed differences in performance in the current study. The dedicated SMwI used a higher‐resolution oblique‐coronal 3D multi‐echo GRE sequence (0.5 × 0.5 × 1.0 mm³), aligning the slice orientation with the long axis of the substantia nigra. In contrast, the SWI‐driven SMwI was acquired in the AC–PC plane with 2‐mm slices, and the NM‐driven SMwI had coarser in‐plane resolution (0.7 × 0.7 × 1.2 mm³). Kim et al. [34] have demonstrated that minimizing partial volume effects requires the use of an oblique coronal plane tangential to SN, along with an appropriate slice thickness, given that the height of SN is approximately 15 mm. However, Lee et al. [35] have suggested that acquisition plane alone may not yield statistically significant differences in diagnostic performance of SMwI, but this does not necessarily imply that plane orientation is irrelevant to partial‐volume effects. From an anatomical and geometric perspective, the dorsolateral nigral hyperintensity extends predominantly along an axis that is approximately parallel to the AC/PC plane. As a result, axial acquisitions aligned with the AC/PC plane are inherently prone to through‐plane averaging of this elongated structure, particularly when relatively thick slices are used. Oblique‐coronal acquisitions, by contrast, sample the nigral hyperintensity in an orthogonal direction, thereby reducing effective partial‐volume contamination even with similar nominal slice thickness. We therefore posit that the impact of acquisition plane on diagnostic performance may be most pronounced through its interaction with slice thickness, rather than as an isolated factor. In the present study, the combination of axial orientation and thicker slices in SWI‐driven SMwI likely exacerbated partial‐volume effects, contributing to the observed loss of specificity.

Consistent with these acquisition differences, reconstructed SMwI exhibited divergent error profiles (Figure 3). SWI‐driven SMwI showed reduced specificity and PPV for all raters, reflecting a tendency toward false‐positive interpretations, likely related to blurred margins of the dorsolateral substantia nigra. By contrast, the NM‐driven SMwI showed lower sensitivity and markedly reduced NPV for all raters, consistent with diminished detection of true nigral hyperintensity loss owing to limited spatial resolution and suboptimal contrast mechanisms. NM imaging has traditionally been performed using Turbo Spin Echo‐based techniques or magnetization transfer (MT)–prepared GRE sequences to enhance neuromelanin‐related contrast [36, 37, 38]. However, simple 3D multi‐echo GRE–based NM imaging has also been validated as a robust alternative, offering high signal‐to‐noise ratio, improved spatial resolution, and vendor‐neutral implementation [27]. Ji et al. [27] demonstrated that GRE‐based approaches can reliably depict neuromelanin‐rich structures while providing superior reproducibility across scanners. In the present study, we deliberately employed a simple 3D GRE–based neuromelanin sequence to reflect a pragmatic, widely deployable clinical protocol rather than a highly specialized or vendor‐specific implementation. Nevertheless, we acknowledge that differences in contrast mechanisms between MT‐prepared and non–MT‐prepared NM imaging may influence the conspicuity of nigral signal changes, which could partly contribute to the reduced sensitivity observed for NM‐driven SMwI. This consideration further underscores that the performance differences observed in this study reflect not only reconstruction strategy but also intrinsic properties of the source imaging sequence.

FIGURE 3.

Error profiles by reconstruction. (a) False positive interpretation on SWI‐driven SMwI of a 67‐year‐old female disease control subject. Arrowheads indicate the dorsolateral substantia nigra, where contrast blurring and ill‐defined margins create the appearance of nigral hyperintensity loss, which is not observed on the standard SMwI, marked with arrow, or supported by the PET scan. (b) False‐negative interpretation on NM‐driven SMwI and an equivocal finding on SWI‐driven SMwI in a 68‐year‐old male patient with Parkinson's disease. Arrow in NM‐driven SMwI indicates preserved‐appearing nigral hyperintensity despite underlying disease, likely reflecting reduced spatial resolution and diminished susceptibility contrast. These error patterns are consistent with the specificity loss observed in SWI‐driven SMwI and the sensitivity loss observed in NM‐driven SMwI in the main analyses.

Across raters, NM‐driven SMwI demonstrated AUC values comparable to those of the standard method, with no statistically significant differences observed for raters 1 and 3, whereas SWI‐driven SMwI consistently showed lower AUC across all raters. The AUC findings should be interpreted alongside sensitivity and specificity, as AUC alone may not fully capture differences in diagnostic performance. Rater‐level differences in AUC precision, with wider confidence intervals for raters 1 and 3, likely reflect expected inter‐rater variability inherent to qualitative assessment rather than systematic bias and further support the use of majority‐vote analysis to improve robustness.

Beyond reconstruction‐based approaches, emerging fast acquisition strategies such as compressed sensing–accelerated SWI/QSM and EPI‐based susceptibility imaging have been proposed to reduce scan time [39, 40, 41, 42]. However, these techniques are not yet widely implemented in routine clinical protocols, and their reliability for depicting fine nigral structures remains under active investigation. In this context, reconstruction‐based SMwI from routinely acquired sequences represents a pragmatic interim solution for clinical settings where dedicated high‐resolution SMwI is not feasible.

From a clinical perspective, these findings support the dedicated SMwI as the preferred protocol for early Parkinson's disease, in which maximizing diagnostic certainty is important. Nevertheless, in the ¹⁸F‐FP‐CIT PET subgroup, SWI‐driven SMwI approached the standard protocol across patient‐based and side‐based analyses, with no significant differences in sensitivity, specificity, or accuracy. However, the PET‐referenced subgroup included relatively few PET‐negative controls, and specificity estimates for the standard SMwI therefore had wide confidence intervals and may be sensitive to rater thresholds. Accordingly, although these PET‐referenced findings should be interpreted with caution, SWI‐driven SMwI may represent a reasonable fallback option when dedicated SMwI cannot be acquired, provided that slice thickness and angulation are carefully optimized. In contrast, the generally inferior performance of NM‐driven SMwI suggests that it is not suitable as a substitute for standard SMwI in its current form.

4.1. Limitations

The SMwI methods compared differed in in‐plane resolution, slice thickness, acquisition plane, and other acquisition parameters including TE, TR, flip angle, and number of echoes, which limits the mechanistic isolation of individual factors and the interpretability of the results. This reflects the pragmatic goal of testing substitution feasibility using sequences already embedded in clinical workflows. Second, the inter‐rater agreement was lower for NM‐driven SMwI, potentially reflecting both technical limitations and differences in rater experience. Third, this was a single‐center retrospective study with a relatively small disease‐control cohort, which may limit generalizability of the findings. Also, inherent biases related to the retrospective nature of the study cannot be entirely excluded. Lastly, the relatively small number of disease control subjects may have limited statistical power for detecting differences in specificity, particularly as reflected by the width of confidence intervals. Accordingly, specificity‐related comparisons should be interpreted with caution, and the observed trends warrant confirmation in larger, more balanced cohorts.

5. Conclusion

Reconstructed SMwI from SWI sequence yielded performance comparable to standard SMwI when referenced to ¹⁸F‐FP‐CIT PET for differentiating early Parkinson's disease, whereas NM‐driven SMwI showed inferior performance in most analyses. However, with specificity trade‐off, SWI‐driven SMwI did not match the diagnostic performance of the dedicated standard SMwI when referenced to clinical diagnosis.

Funding

This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government (MSIT) (grant numbers: RS‐2026‐25468417 and RS‐2023‐00237618) and the Korea Health Technology R&D Project through the Korean Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: RS‐2023‐00266288).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Demographic characteristics regarding ¹⁸F‐N‐3‐fluoropropyl‐2‐β‐carbomethoxy‐3‐β‐(4‐iodophenyl)nortropane PET.

Table S2: Diagnostic performance of standard, NM‐driven and SWI‐driven SMwIs using clinical assessment as reference standard for rater 1.

Table S3: Diagnostic performance of standard, NM‐driven and SWI‐driven SMwIs using clinical assessment as reference standard for rater 3.

Table S4: Diagnostic performance of standard, NM‐driven and SWI‐driven SMwI methods in the ¹⁸F‐FP‐CIT PET subgroup for rater 1. A total of 106 subjects (98 ePD and 8 DC) were included in the subgroup analysis.

Table S5: Diagnostic performance of standard, NM‐driven and SWI‐driven SMwI methods in the ¹⁸F‐FP‐CIT PET subgroup for rater 3. A total of 106 subjects (98 ePD and 8 DC) were included in the subgroup analysis.

Data Availability Statement

The datasets generated or analyzed during the study are available from the corresponding author on reasonable request.