Truace
Subscribe

In 267 prostate cancer patients, AI-measured tumour volume on scans tracked survival

Study title: Fully automated AI-based quantification of tumour volume on PSMA PET-CT images is associated with overall survival in prostate cancer

A Swedish retrospective study followed 267 patients who had [18F]PSMA-1007 PET-CT at Skåne University Hospital for primary staging or biochemical recurrence of prostate cancer. A fully automated AI model measured tumour volume without manual corrections. During five years of follow-up, 48 patients (18%) died, and higher whole-body tumour volume was associated with higher mortality (p = 0.001).

Read the study abstract

Introduction Artificial intelligence (AI)-based models can quantify tumour volume in positron emission tomography (PET) examinations using prostate specific membrane antigen (PSMA). It is, however, unknown whether this measurement of tumour volume contains prognostic information. The purpose of this study was to examine whether tumour volume measured by AI is associated with overall survival in patients with prostate cancer. Methods Patients who had undergone [ 18 F]PSMA-1007 PET-CT due to primary staging or bio…

Truvace record2 min read1 sourceEvidence: Moderate · 60/100 — Observational study · n = 267 · 1 independent sourcePeer-reviewedTRV-2026-2414 · cite or check it
Fully automated AI-based quantification of tumour volume on PSMA PET-CT images is associated with overall survival in prostate cancer
Gallium PSMA PET scan by Rauscher, Isabel; Maurer, Tobias; Fendler, Wolfgang P.; Sommer, Wieland H.; Schwaiger, Markus; Eiber, Matthias. CC BY 4.0 · https://creativecommons.org/licenses/by/4.0

In brief

Patients with high whole-body volume had a hazard ratio of 3.20 (95% CI 1.50–6.80) compared with the low group. The link was significant in the 159-patient primary staging group (p = 0.014) but did not reach significance in the 108-patient recurrence group (p = 0.061). The authors say this suggests scans could yield prognostic information without the time cost of manual outlining.

The result is an association in one centre's retrospective data, and the authors say further validation is needed before clinical use. Unreviewed AI output included false positives, which may blur the categories, especially for patients with low tumour burden.

Main points

  1. The cohort had 267 patients: 159 (60%) scanned for primary staging and 108 (40%) for biochemical recurrence, with a median age of 71.
  2. Whole-body tumour volume was split into tertiles; the high group had a hazard ratio of 3.20 (95% CI 1.50–6.80) versus the low group, adjusted for age.
  3. Bone tumour volume was also linked to survival (p = 0.016), with a hazard ratio of 2.20 for high versus low.
  4. Prostate and lymph node volumes did not reach overall significance (p = 0.063 and p = 0.071), though high groups had higher mortality.
  5. In the primary staging cohort, 29 patients (18%) died; whole-body volume was significant (p = 0.014) with a hazard ratio of 3.99 (95% CI 0.93–17.20).
  6. In the recurrence cohort, 19 patients (18%) died; whole-body volume did not reach significance (p = 0.061) despite a hazard ratio of 3.47.
  7. None of the scans were part of the AI model's training data.

The gain

In a retrospective study of 267 prostate cancer patients, tumour volume measured automatically by AI on PSMA PET-CT, with no manual correction, was linked to survival, and high whole-body volume carried a hazard ratio of 3.20 for death versus low.

The full story

Patients were consecutively invited between September 2019 and July 2021 and scanned on a Discovery MI PET-CT two hours after injection of the tracer. The AI model segmented lesions in the prostate or prostate bed, lymph nodes and bone, and volumes were summed into total lesion volume. Mortality was compared with Cox regression adjusted for age.

The AI assigned bone tumour volume above zero to about 60% of patients, versus roughly 20% with manual outlining in an earlier, largely overlapping cohort. The authors say small or non-specific bone findings were counted, so AI bone volume should not be read as the exact volume of bone metastases. Positive predictive value for suspected bone metastases was about 70%, with 0.76 false-positive lesions per patient.

Patients with small but measurable prostate or lymph node volume tended to have lower mortality than those with zero measured volume, in both the AI and manual studies. The authors suggest selection bias, since absent prostate disease may reflect spread elsewhere, and low PSMA expression in 5–10% of cancers as possible explanations.

Sources

  1. Peer-reviewedClinical Physiology and Functional Imaging2026-11-01

The debate