LaVCa: LLM-assisted Visual Cortex Captioning