Automated Tumor Segmentation, Identification, Dosimetry and Response Assessment for Lu-177 Radiopharmaceutical Therapy
Abstract
Purpose
To develop an automated tool for tumor segmentation, identification, dosimetry, and response assessment for radiopharmaceutical therapy (RPT).
Methods
22 patients treated with Lu-177 RPT (15 Pluvicto, 7 Lutathera) were analyzed retrospectively. Quantitative SPECT/CTs were acquired ~24 hours post-injection for 3D dosimetry using the Hänscheid Single-Time-Point formulism. Tumors were automatically segmented from SPECT/CT using SUV thresholds of 3 for body and 2 x background for liver, excluding physiological activities. The locations of tumors were automatically identified using deep-learning-based auto-contours. Mean doses to individual tumors were corrected for the time of SPECT imaging based on published population data and partial volume effect. Response was assessed for 9 Pluvicto patients who were scanned after 2 cycles. Tumors from cycle 1 were registered and deformed to cycle 2, then updated with SUV 3 threshold. Tumor burden (volume x SUVmean) was used to track responses, with >=20% reduction defined as responding and >=20% increase defined as progression.
Results
424 tumors were segmented from 31 SPECT/CT scans. Location was successfully identified for 88% of tumors. For 9 patients who were scanned after 2 cycles of injections, 7 were responders (total tumor burden reduced 28%-95%), and the other 2 were stable (= 5.5 Gy, the response rate was 90%.
Conclusion
Using quantitative SPECT/CT, an automated tumor segmentation, identification, dosimetry, and response assessment tool was developed for radiopharmaceutical therapy. Tumor response is dose- and location-dependent.