Giacomo De Rosa

Giacomo De Rosa

Auto-generated profile, based on publicly available information

Giacomo De Rosa


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
29 April 2025 in JMIR Preprints
Verified
Preprint
Open access
GOGiovanni Ottonelli
Giovanni Ottonelli
GRGiacomo De Rosa
Giacomo De Rosa
JBJacopo Celada Ballanti
Jacopo Celada Ballanti
AGAlessandro Gaeta
Alessandro Gaeta
PVPaolo Vinciguerra
Paolo Vinciguerra
Alessandra Di MariaAlessandra Di Maria
Alessandra Di Maria
BACKGROUND Intraocular lens (IOL) power calculation is crucial for achieving optimal refractive outcomes in cataract surgery. Several formulas have been developed to predict postoperative refraction, with varying degrees of accuracy. Recent advancements in artificial intelligence (AI) have led to the exploration of AI-driven tools as potential alternatives or adjuncts to traditional formulas. DeepSeek, a generative AI model, is one such tool that has been proposed for biometric calculations. However, its performance in predicting refractive outcomes compared to conventional methods remains unclear. This study aims to assess the accuracy of DeepSeek in predicting postoperative refraction and compare it with a widely used formula, the Barrett Universal II. OBJECTIVE To evaluate the accuracy of DeepSeek, an open-source generative artificial intelligence (AI), in predicting postoperative refractive outcomes for intraocular lens (IOL) power calculation compared to the Barrett Universal II formula. METHODS This study analyzed biometric data from 100 eyes of 50 patients who underwent cataract surgery between July 2024 and January 2025 at Humanitas Research Hospital in Milan, Italy. Only uncomplicated cases of emmetropia with Alcon AcrySof® SA60WF IOL implantation were included. Postoperative objective refraction was measured with Nidek TONOREF III. Predicted spherical equivalents were obtained using the Barrett Universal II formula and DeepSeek AI. Mean refractive error (MRE), standard deviation (SD), and cumulative frequency of target refractive error were calculated. A Wilcoxon signed-rank test was performed to assess statistical significance. RESULTS The MRE was significantly lower for Barrett (0.262 D) than for DeepSeek (0.659 D) (P < 0.0001). The SD was 0.386 D for Barrett and 0.424 D for DeepSeek. Barrett achieved refractive accuracy within ±0.25 D in 37% of cases, ±0.50 D in 75%, ±0.75 D in 85%, and ±1.0 D in 90%. DeepSeek performed worse at each threshold, reaching 10%, 30%, 55%, and 80%, respectively. CONCLUSIONS DeepSeek AI demonstrated significantly higher refractive errors and lower predictive accuracy than the Barrett Universal II formula. Although AI is a valuable tool already integrated into some biometric formulas, current widely accessible generative models remain inadequate for accurate biometric calculations, emphasizing the need for human-validated formulas to ensure optimal refractive outcomes in cataract surgery.