**Background:** Selective retina therapy (SRT) is a minimally invasive laser treatment that targets the retinal pigment epithelium (RPE) while sparing the neurosensory retina. It relies on microbubble formation (MBF) around melanosomes within the RPE, induced by microsecond laser pulses. However, the optimal pulse duration and application mode to avoid cumulative thermal damage and ensure safety remain unclear. Real-time feedback dosimetry (RFD) is essential to adjust laser energy in response to variable RPE pigmentation and ocular transparency. This study investigated the influence of pulse duration (8, 12, 16, 20 µs) and application mode (single, ramp, burst) on RPE damage thresholds and evaluated an SD-OCT-based RFD algorithm.
**Methods:** Nine ex vivo porcine eyes were used. A prototype SRT laser (532 nm, 30 W, 90×90 µm² spot) delivered pulses at 100 Hz. A total of 1620 lesions were applied in the visual streak area. RPE damage was assessed using a live/dead fluorescence assay (calcein-AM/ethidium homodimer-1) with two criteria: (1) “area” – lesion area >50% of the 90×90 µm² spot; (2) “cluster” – ≥3 contiguous dead cells. Suprathreshold lesions were evaluated by color fundus photography (CFP) and OCT B-scans. Time-resolved OCT M-scans (85 kHz, 870 nm) were recorded simultaneously for RFD. Fringe washouts (signal loss ≥1 A-scan integration time of 11.8 µs and ≥5 pixels axially) were detected visually and by an algorithm with adjustable κ-values (5–100). Probit analysis yielded ED₁₆, ED₅₀, and ED₈₄ values. Statistical performance (sensitivity, specificity, accuracy, AUC) was calculated against the viability assay.
**Key Results:** For pulse durations ≤12 µs, no cumulative RPE damage was observed in ramp mode (threshold ratio single/ramp ≈1.0). For 16 and 20 µs, cumulative effects appeared: ramp mode lowered the ED₅₀ by a factor of 1.2, and burst mode by a factor of 1.5 (area criterion). For example, at 20 µs, single-pulse ED₅₀ was 126 µJ (1555 mJ/cm²) vs. burst ED₅₀ of 87 µJ (1074 mJ/cm²). At 8 µs, no threshold difference was seen among modes. CFP detected suprathreshold lesions only in 2 of 9 samples; in one sample (No. 6), the therapeutic window (TW) was 2.1 and safety range (SR) 1.9 for 8 µs pulses. The RFD algorithm achieved its best performance for 8 µs pulses in ramp mode: sensitivity 96%, specificity 97%, AUC 0.979 (κ=15–17). For single pulses at 8 µs, sensitivity 96%, specificity 96%, AUC 0.983 (κ=18). Across all samples and settings, a median κ=12 gave overall accuracy 93% (sensitivity 91%, specificity 94%). Investigator-based evaluation showed highest accuracy (97%) for 8 µs single pulses (sensitivity 98%, specificity 90%) but poor specificity for longer pulses (e.g., 16 µs single: accuracy 50%, specificity 34%).
**Clinical Implications:** The study demonstrates that microsecond laser pulses shorter than 12 µs are safe for SRT because they avoid cumulative RPE damage, even with multiple pulses at 100 Hz. Longer pulses (16–20 µs) risk heat accumulation and are less suitable unless tightly controlled by RFD. The SD-OCT-based RFD algorithm, particularly with 8 µs pulses in ramp mode, provides high sensitivity and specificity for detecting RPE damage, enabling real‑time dose adjustment to stay within the therapeutic window. This could improve the safety and efficacy of SRT for conditions like central serous chorioretinopathy and diabetic macular edema. The algorithm’s performance in ex vivo eyes suggests even better results in vivo due to improved OCT signal quality. Future clinical studies (e.g., NCT04968756) will validate these findings in patients.