Table 7 Topic-wise accuracy comparison with training sample distribution. Bold values indicate the best-performing accuracy within each topic between the two models.
From: Ophtimus-V2-Tx: a compact domain-specific LLM for ophthalmic diagnosis and treatment planning
Topic | Case report, % | MCQA, % | Ophtimus-V2-Inst | Ophtimus-V2-Tx |
|---|---|---|---|---|
Anterior segment | 0.35 | 0.07 | 0.53 | 0.40 |
Cataract | 4.08 | 5.76 | 0.50 | 0.56 |
Conjunctiva | 3.03 | 3.64 | 0.80 | 0.57 |
Cornea | 13.76 | 11.76 | 0.46 | 0.46 |
Error of refraction | 0.83 | 3.62 | 0.68 | 0.26 |
General ophthalmology | 0.70 | 4.83 | 0.69 | 0.46 |
Glaucoma | 6.38 | 7.57 | 0.59 | 0.42 |
Neuro ophthalmology | 16.26 | 15.14 | 0.78 | 0.40 |
Ocular trauma | 3.14 | 1.85 | 0.82 | 0.42 |
Oculoplastic | 1.37 | 2.04 | 0.45 | 0.54 |
Optics | 0.02 | 4.09 | 0.38 | 0.25 |
ORBIT_LIDS_ADNEXA | 1.56 | 1.39 | 0.60 | 0.47 |
Pathology | 2.86 | 1.77 | 0.61 | 0.47 |
PEDIATRICS_STRABISMUS | 2.68 | 8.60 | 0.63 | 0.38 |
PHARMACOLOGY | 0.35 | 3.34 | 0.75 | 0.55 |
POSTERIOR SEGMENT | 2.73 | 0.63 | 0.57 | 0.26 |
Retina and vitreous | 26.14 | 13.63 | 0.46 | 0.46 |
Systemic diseases | 4.37 | 2.21 | 0.87 | 0.62 |
UVEITIS | 8.57 | 6.16 | 0.72 | 0.80 |
Others | 0.83 | 1.92 | – | – |
Average accuracy | – | – | 0.62 | 0.47 |
Total | 9269 | 43,983 | – | – |
