Reply to Koulaouzidis and Marlicz

10.1055/a-2808-4380

We thank the authors for their thoughtful and constructive response to the study by Thoufeeq et al., which evaluated interobserver and intra-observer agreement in Size, Morphology, Site, and Access (SMSA) scoring [1]. The findings, particularly the only fair agreement for overall SMSA level and the poor concordance in size estimation, carry direct and clinical implications for referral pathways, therapeutic planning, and risk stratification. Similar findings have been reported even with the Paris classification, which continued to demonstrate only fair to moderate interobserver agreement, even among expert endoscopists, underscoring the inherent variability of visually based morphological assessment frameworks [2].

We are grateful to A. Koulaouzidis and W. Marlicz for emphasizing a critical and persistent challenge: polyp size estimation and characterization remain unreliable, even among experienced endoscopists. Their observation that expert visual assessment may be an insufficient gold standard for development of artificial intelligence (AI) tools is particularly pertinent. At the same time, it is important to recognize that some new AI-based size estimation often relies on computational techniques that do not use expert visual judgement as the reference standard [3].

We agree with A. Koulaouzidis and W. Marlicz that our study reinforces the need to refine existing scoring systems before meaningful automation can be achieved. In this context, we would like to offer several additional considerations.

First, reducing reliance on subjective size estimation, either by down-weighting size within composite scores or by adopting broader size categories that may help minimize misclassification around clinically relevant thresholds. Second, existing scoring systems should be rebalanced to prioritize parameters with higher reproducibility, and established frameworks such as SMSA should be adapted appropriately for only diagnostic modalities. Wherever feasible, objective standards should be incorporated. Third, training and reporting would benefit from greater standardization, including use of video-based reference libraries and agreed benchmarks.

We also suggest that AI be introduced selectively, targeting domains where human performance is demonstrably weakest, particularly size measurement, while avoiding overreliance on subjective human labels during model development. Furthermore, defining acceptable limits of human variability and aligning scoring thresholds with clinical decision-making may help ensure that minor disagreements do not translate into major downstream consequences.

Finally, it remains essential to acknowledge that, despite its promise, AI in colonoscopy is not yet ready for widespread adoption. Several limitations and unresolved challenges must be addressed before integration into routine clinical practice can be justified.

Publication History

Received: 25 January 2026

Accepted: 06 February 2026

Article published online:
27 February 2026

• References

• 1 Thoufeeq M, Thaika A, Moudhgalya S. et al. Is size, morphology, site, and access scoring system consistent between endoscopists? Interobserver and intraobserver polyp assessment study. Endosc Int Open 2025; 13: a27522591
• 2 Djinbachian R, Alj A, Medawar E. et al. Interobserver agreement for the Paris classification of colorectal lesions amongst surgeons, gastroenterologists, trainees and experts: A video-based study. Dig Dis Sci 2025; 70: 4122-4129
• 3 Yii CY, Toh DE, Chen TA. et al. Use of artificial intelligence to measure colorectal polyp size without a reference object. Endosc Int Open 2025; 13: a25561836