Table 5 Ablation study of updated key components in the PlaqueCap framework on the IVUS-Lesion-Caps dataset

From: PlaqueCap: lesion-centered captioning of atherosclerotic plaques in intravascular ultrasound using vision-language models and prompt injection

Model Variant

BLEU-4

CIDEr

PlaqueCap (Full Model)

33.8

82.4

w/o Segmentation Guidance

27.5

70.3

w/o Lesion Prompt Injection (LPI)

29.8

74.1

Global Pooling Baseline

30.5

75.8

w/o Prompt Mixer

31.6

77.0

Simple Additive Injection (vs. LoRA)

32.5

79.5