Thyroid Cancer Staging with SPECT: Current Limitations and AI Solutions

Differentiated thyroid cancer - papillary and follicular types - is the most common endocrine malignancy and one of the few cancers where radioiodine therapy provides both staging information and therapeutic effect in a single administration. I-131 post-treatment whole-body scanning and SPECT/CT imaging map residual thyroid tissue, locoregional recurrence, and distant metastases with functional specificity that anatomical imaging cannot match. The limitation that has persisted through decades of clinical use is the spatial resolution of I-131 SPECT - typically 12-15 mm FWHM, substantially lower than Tc-99m imaging due to the 364 keV energy of I-131 requiring high-energy collimators. AI reconstruction is changing what that limitation means in practice.

The Diagnostic Stakes in Thyroid Cancer Restaging

Post-thyroidectomy I-131 treatment is followed by a post-treatment scan 5-7 days after administration, when the therapeutic activity remaining in the patient is sufficient for imaging. This scan serves two purposes: confirming thyroid remnant ablation and detecting unexpected disease sites that were not identified on pre-treatment diagnostic imaging. The second purpose is the clinically consequential one - when the post-treatment scan identifies distant metastases not previously known, the patient's risk classification and subsequent management change substantially.

ATA (American Thyroid Association) risk stratification distinguishes low, intermediate, and high recurrence risk based on pathological features, surgical extent, and post-treatment scan findings. A patient classified as intermediate risk who shows unexpected lung metastases on the post-treatment SPECT moves immediately to high-risk management - including consideration of repeat high-dose I-131 treatment, alternative systemic therapies, and more intensive surveillance imaging. This reclassification happens based on SPECT findings in approximately 8-12% of patients in published series who had intermediate-risk pathological features.

The consequence of failing to detect a pulmonary metastasis on the post-treatment SPECT - either because the lesion is below the resolution threshold or because it is obscured by residual thyroid bed uptake - is a delayed reclassification at first surveillance imaging 6-12 months later. That delay does not change disease biology in most cases, but it delays initiation of appropriate management and increases patient anxiety during the surveillance period.

I-131 SPECT Resolution: The Physics of High-Energy Imaging

The spatial resolution deficit in I-131 SPECT compared to Tc-99m SPECT is a direct consequence of collimator physics. Higher-energy photons (364 keV for I-131 versus 140 keV for Tc-99m) require heavier lead septa in the collimator to prevent penetration - which requires wider holes to maintain acceptable sensitivity. The resolution-sensitivity tradeoff is less favorable at higher energies: a high-energy parallel-hole collimator achieving adequate sensitivity for I-131 imaging provides system resolution of approximately 12-15 mm FWHM at 10 cm source distance, versus 8-10 mm FWHM for Tc-99m imaging with a LEHR collimator.

This means that a pulmonary metastasis must be approximately 18-25 mm in diameter to appear with adequate contrast for reliable detection on I-131 SPECT - the partial volume threshold at 2-2.5x the system resolution. Miliary pulmonary metastases, which are common in follicular thyroid cancer and represent a favorable prognostic subtype that often responds well to repeated I-131 treatment, frequently fall below this threshold and may be invisible on SPECT even when clearly detectable on CT.

CT-SPECT fusion on modern hybrid SPECT/CT cameras helps anatomical localization but does not resolve the underlying sensitivity limitation. A lesion that generates insufficient counts in the SPECT data due to its small size and the camera's resolution limitations will not be recovered by fusion with the CT - the CT shows the lesion anatomically, but without concordant SPECT uptake, it may be attributed to a non-iodine-avid etiology and not acted upon.

Deep Learning Reconstruction: Resolution Recovery Without Artifact

Deep learning-based reconstruction for I-131 SPECT is technically more challenging than for Tc-99m-based studies. The higher energy creates more significant septal penetration and scatter contributions, the count rates are lower for a given administered activity, and the image noise characteristics differ from lower-energy acquisitions. Models trained on Tc-99m data do not transfer to I-131 without domain adaptation.

Specialized I-131 SPECT reconstruction networks, trained on Monte Carlo-simulated data that accurately models high-energy scatter and collimator penetration physics, show resolution recovery equivalent to approximately 40% reduction in FWHM - bringing effective resolution from 13-15 mm down to 8-9 mm. In phantom validation, this allows reliable detection of lesions of 10-12 mm diameter versus the 18-20 mm threshold for conventional reconstruction.

The clinical translation of this phantom performance to patient studies is the critical validation step. Published patient studies with I-131 deep learning reconstruction, compared to conventional HEUV collimator reconstruction, show additional lesion detection in 18-22% of cases - mostly in the lungs, cervical lymph nodes, and bone. The additional lesions detected are predominantly small (<15 mm) and represent the population most likely to be missed on conventional reconstruction.

SPECT vs. PET for Thyroid Cancer: A Misunderstood Comparison

The oncology community sometimes frames thyroid cancer imaging as a choice between I-131 SPECT/CT and I-124 PET/CT. I-124, the positron-emitting iodine isotope, provides true PET resolution (4-5 mm) for pre-treatment dosimetry and disease staging. Where I-124 is available - primarily at academic centers with cyclotrons - it provides significantly higher sensitivity for small pulmonary metastases and detailed pre-treatment lesion characterization.

The practical reality is that I-124 PET/CT is available at fewer than 30 centers in the United States. The post-treatment I-131 scan, performed with the therapeutic activity, is available at every institution that performs thyroid cancer treatment - approximately 800 hospitals. Improving I-131 SPECT diagnostic performance through deep learning reconstruction is therefore the intervention with the widest potential impact on thyroid cancer staging accuracy in the actual patient population being treated, not just in academic center referrals.

As we discuss in our overview of deep learning for SPECT imaging, the argument for investing in SPECT reconstruction quality applies across multiple indications where PET access is limited - thyroid cancer is among the clearest cases where the gap between available technology and best available technology creates direct patient impact.

Quantitative Dosimetry in Thyroid Cancer Treatment

I-131 dosimetry for thyroid cancer uses the same framework as RPT dosimetry generally: serial imaging at multiple time points, organ segmentation, time-activity curve fitting, and S-value-based dose calculation. For thyroid cancer, the organs of interest are the thyroid remnant (when ablation is the goal), locoregional disease, and distant metastases. Bone marrow dosimetry is performed for patients receiving high activities to ensure the 2 Gy marrow dose constraint is not exceeded.

Automated dosimetry for thyroid cancer is relevant to the subgroup of patients with significant metastatic disease receiving repeated high-dose I-131 treatment - a clinically important but relatively small fraction of the overall thyroid cancer population. For routine ablation, the fixed-activity approach (typically 30-150 mCi depending on risk category) remains standard and does not require individualized dosimetry. The precision dosimetry approach is most impactful in the high-burden metastatic setting where maximizing delivered dose while protecting marrow drives treatment planning decisions.