Digitizing paper-based medical reports containing both printed and handwritten text is vital for integrating legacy records into modern electronic health systems, yet this task remains hindered by degraded paper quality, handwriting variability, and complex medical terminology. This study presents a framework that integrates Optical Character Recognition (OCR) with Multimodal Large Language Models (MLLMs)—including DeepSeek, ChatGPT (GPT-4V), and Gemini—for contextual understanding and clinical interpretation. Psoriasis was chosen as a representative case due to its longitudinal, detailed documentation encompassing Psoriasis Area and Severity Index (PASI) scores, treatment histories, and lesion imagery. Using 360,000 handwritten psoriasis reports from 4,500 patients, the system accurately extracts and structures key clinical information, including medications, disease progression, and treatment outcomes. Unlike prior OCR-LLM efforts, this work introduces a multimodal integration of textual and lesion image data, producing structured JSON outputs compatible with electronic health record (EHR) systems for practical clinical deployment. Model performance, evaluated using Character Error Rate (CER), Word Error Rate (WER), and F1-score, demonstrated Gemini's superior accuracy across all metrics, establishing a strong baseline for future multimodal training on domain specific medical datasets.