# Investigating AD and non-AD [18F]flortaucipir distribution patterns in a memory clinic cohort.

- **DOI:** [10.1007/s00259-026-07995-z](https://doi.org/10.1007/s00259-026-07995-z)
- **Rating:** Limited (2.5 out of 5)
- **Verdict:** A useful memory-clinic replication shows tau-PET patterns can separate A+T+ cases, but the design does not establish independent amyloid-status prediction.
- **Framework version:** v5.9
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-10-08
- **Evaluation page:** https://www.nabu.science/paper/10.1007%2Fs00259-026-07995-z

## Key claims

1. [18F]flortaucipir-PET DPs derived from SSM/PCA provide information on both A and T status
   - Support: Tentative
   - Type: Primary finding
   - Basis: the design mainly distinguishes A+T+ from A−T± and does not independently identify amyloid positivity
   - Location: Conclusion
2. all machine learning models confirmed the capacity of the A+DP to predict AD pathology, correctly classifying at least 36 out of 40 cases
   - Support: Tentative
   - Type: Secondary
   - Basis: the reported count is unreconciled with the multiclass metrics and arises from label-informed same-cohort evaluation
   - Location: Abstract Results
3. the A−DP showed limited discriminative power for non-AD conditions
   - Support: Supported
   - Type: Secondary
   - Basis: the direction is consistent across reported overlap and classifier results despite the small heterogeneous 13-person subgroup
   - Location: Abstract Results; Discussion
4. Their representativeness was further supported by significant correlations with plasma biomarkers in nearly all cases
   - Support: Tentative
   - Type: Secondary
   - Basis: correlations use only 43 participants and include two explicitly non-significant A−DP comparisons
   - Location: Abstract Results; Results §Correlation Between SSM/PCA Scores and Plasma Biomarkers
5. The MLP achieved the highest overall performance (balanced accuracy = 0.81, weighted-F1 = 0.87, macro-F1 = 0.81).
   - Support: Tentative
   - Type: Secondary
   - Basis: the point estimates are reported, but the analytical denominator and same-dataset validation limit their reliability
   - Location: Results §Amyloid Status Prediction

## Strengths

- **Clinically relevant heterogeneous replication cohort** The analysis tests earlier SSM/PCA findings in a memory-clinic cohort spanning cognitive stages and includes a heterogeneous A−T+ group, exposing limitations that a narrowly defined non-AD cohort may miss.
  - Dimension: contribution
  - Location: Methods §Study Cohort; Discussion, comparison with Hammes et al.
- **Detailed multistage imaging analysis pipeline** The Methods describe image processing, SSM/PCA construction, LOOCV subject scoring, three tuned classifiers and imbalance-robust metrics. A repeated undersampling sensitivity analysis further examines class imbalance.
  - Dimension: methodological rigour
  - Location: Methods §Multivariate Analysis and §Amyloid Status Prediction; Results §Amyloid Status Prediction
- **Candid negative non-AD finding** The Results and Discussion report substantial overlap and limited discrimination for the A− pattern, then relate this to the etiological heterogeneity of the A−T+ group.
  - Dimension: positioning
  - Location: Results §Multivariate Analysis and §Amyloid Status Prediction; Discussion, A−T+ heterogeneity paragraph

## Limitations

- **Cohort denominator and target remain inconsistent** The Abstract states n=81, while AT Profiles Computation and Table 1 include 17 additional A+T− participants. The classifier defines A+T+ rather than all amyloid-positive participants as positive, obscuring which population and target the reported performance represents.
  - Dimension: methodological rigour
  - Location: Abstract; Methods §AT Profiles Computation and §Amyloid Status Prediction; Table 1
- **Headline performance counts lack reconciliation** The Abstract's “36 out of 40” and “7 out of 13” counts are not presented in the Results text or reconciled with the multiclass balanced accuracy and F1 metrics. This is a material internal-coherence concern for the central performance claims.
  - Dimension: reporting
  - Location: Abstract Results; Results §Amyloid Status Prediction; Figure 4
- **Clinical conclusion exceeds demonstrated inference** The Discussion states that apparent A+ identification is driven by the cohort's association between T+ and A+, yet the Conclusion proposes tau PET as a possibly sole examination for suspected AD. A+T− scores resembling A−T− further limit the amyloid-specific interpretation.
  - Dimension: positioning
  - Location: Results §Multivariate Analysis; Discussion, classifier interpretation paragraph; Conclusion

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Noted
  - Their representativeness was further supported by significant correlations with plasma biomarkers in nearly all cases: "Significant correlations were observed across all comparisons (p < 0.01)." (Results, Correlation between SSM/PCA scores and plasma biomarkers) vs "No significant correlations were found between the A − DP scores and either Aβ42/Aβ40 (rs = 0.02, p = 0.92) or p-tau231 (rs = 0.30, p = 0.05)." (Results, Correlation between SSM/PCA scores and plasma biomarkers). The five biomarkers paired with two DP scores give ten comparisons, of which the text identifies two as nonsignificant, so the universal statement must exclude those two while the key claim of significance in nearly all cases remains consistent. → Conclusion holds
  - Minor reporting notes (1):
    - Descriptive: "A cohort of 81 subjects was selected from an ongoing study at the Geneva Memory Clinic" (Methods, Study cohort) vs "Group nameA − T − (n = 28)A + T − (n = 17)A − T + (n = 13)A + T + (n = 40)" (Results, Table 1 header). The four listed group sizes sum to 28 + 17 + 13 + 40 = 98, whereas the three primary groups excluding the explicitly described auxiliary A+T− test group sum to 81. → Conclusion holds
- **Reporting and transparency:** Noted. The 2D classifier's stated amyloid-status target is not explicitly reconciled with its multiclass output. The Table 1 caption omits the reported age difference, and acquisition details are supplied for florbetapir but not flutemetamol. These are reporting issues, not demonstrated changes to the performance results.
  - Ethics approval: declared · Data availability: not stated · Code availability: not stated
- **Reference integrity:** Noted
  - 34 references in manuscript
  - 34 of 34 checkable references found in an index
  - the reference list may be incompletely parsed; counts are indicative
  - References not verified automatically are resolved manually within 24 hours.

## Post-publication record

- **Status:** Clean
- **Summary:** No retraction or expression of concern found in Crossref (includes Retraction Watch).
- **Last checked:** 2026-10-17
