Automated histomorphometry of ablative fractional laser lesions enables a complete-field census of geometric heterogeneity
Year: 2026
Type: article
Abstract: Ablative fractional laser (AFL) treatments are evaluated histologically by measuring about ten selected lesions per sample, which cannot describe the whole field. We developed an automated framework that segments ablation, coagulation, viable tissue and background on horizontal nitroblue tetrazolium chloride (NBTC)-stained sections and measures every lesion. Six models (five architectures) were trained under a pre-specified protocol and tested on all 35 serial sections of a held-out punch. The final pipeline averages three architectures selected on validation data. On the held-out punch it reached Dice 0.932 for ablation and 0.911 for coagulation, detected 96.3% of lesions with 93.7% precision after a lesion filter (62.4% without), and measured ablation diameter with a mean bias of −0.1 \(\mu\) m. Accuracy decreased with depth (deepest band 0.858). These values come from a single punch that is an outlier in treatment density (pitch 2.1 times nominal); performance on specimens of typical density has not been confirmed. In expert annotations, a random 10-lesion sample estimated within-section size variability within ±20% in a median of 58% of draws, and selecting the 10 best-formed craters underestimated it in every section (median −63%). Complete-field measurement is therefore needed; as donor identity was not recorded, between-sample differences may reflect few donors.
Source: Scientific Reports
Institution: Massachusetts General Hospital
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7225824367
FWCI:
Cites:
Cited by:
Related to: