Treatment response in metastatic prostate cancer is typically assessed at the patient level — is the patient getting better or worse? But individual lesions within the same patient can respond differently to the same treatment. Some shrink. Others grow. Some appear. Knowing which lesions respond and which don't would change treatment decisions.
Note (2026-11-09): The abstract (Yousefirizi, Beauregard and Rahmim) is a review of clinical motivations, technical foundations and future pathways for automated lesion tracking in PSMA PET/CT and post-therapy SPECT/CT, tied to response frameworks such as RECIP 1.0 and to lesion-level dosimetry in 177Lu-PSMA therapy; it says recent AI advances 'now make scalable longitudinal lesion correspondence feasible'. The essay's claim that the bottleneck in precision oncology is the longitudinal matching problem rather than imaging sensitivity, and that modern scanners detect lesions reliably, is not in the abstract.
The authors (arXiv:2603.22666) examine the technical requirements for automated lesion-by-lesion tracking across serial PSMA PET/CT and post-therapy SPECT/CT scans. The challenge is threefold: segmenting lesions accurately in each scan, matching them across time points (which lesion in scan 2 corresponds to which in scan 1?), and handling the differences between PET/CT and SPECT/CT imaging.
Recent AI methods now make this scalable. Automated segmentation handles the hundreds of metastatic sites that manual annotation cannot. Automated matching tracks individual lesions through anatomical changes.
The through-claim: the bottleneck in precision oncology for metastatic disease isn't imaging sensitivity — modern scanners detect individual lesions reliably. It's the longitudinal matching problem: connecting the same lesion across scans taken weeks or months apart, after the patient has moved, the anatomy has changed, and some lesions have appeared or disappeared. Scaling this matching is what converts response assessment from patient-level to lesion-level.