Fig. 2: Sample reports generated by the proposed model (MORG), large language models along with the corresponding ground truths written by ophthalmologists, translated from the original Chinese version. | npj Digital Medicine

Fig. 2: Sample reports generated by the proposed model (MORG), large language models along with the corresponding ground truths written by ophthalmologists, translated from the original Chinese version.

From: A deep learning based automatic report generator for retinal optical coherence tomography images

Fig. 2

Manual scores by two retinal subspecialists are provided for clarity. “w/ ins. “ refers to “with instructions”, while “w/o ins. “ refers to “without instructions”. The reports were graded with a 5−point scale: 1: Very poor/unacceptable error. 2: Poor/minor potentially harmful error. 3: Medium/misinterpreted error. 4: Good/only minor harmless error. 5: Very good/no mistakes.

Back to article page