Scholarlink

Medical clinic location selection critically impacts long-term profitability, yet physicians often lack data-driven decision-making tools for this strategic choice. This research addresses key challenges in healthcare facility planning: the scarcity of revenue prediction data and limited understanding of location-revenue relationships specific to medical clinics. We propose an innovative framework utilizing Large Multimodal Models (LMMs) to extract latent regional characteristics from satellite imagery, enabling effective revenue prediction despite small sample constraints. Our methodology integrates municipal-level demographic data with latent LMM-estimated features, capturing both explicit metrics and implicit characteristics that conventional methods overlook. Experiments across multiple clinic types and location conditions demonstrated that incorporating latent LMM-estimated features consistently improved prediction performance, with average improvements of 8.3% for mean squared error (MSE) and 31.6% for coefficient of determination (R \(^2\) ) compared to demographic-only approaches. The extracted interpretable features provide concrete decision-making criteria for practitioners. This versatile framework extends beyond healthcare to optimize location strategy across various industries facing similar data constraints.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Medical Clinic Revenue Prediction Using Latent Feature Extraction from Satellite Imagery with Large Multimodal Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shuntaro Masuda,
  • Fumiya Matsuno,
  • Itsuki Hirai,
  • Koji Muta,
  • Toshihiko Yamasaki

摘要

Medical clinic location selection critically impacts long-term profitability, yet physicians often lack data-driven decision-making tools for this strategic choice. This research addresses key challenges in healthcare facility planning: the scarcity of revenue prediction data and limited understanding of location-revenue relationships specific to medical clinics. We propose an innovative framework utilizing Large Multimodal Models (LMMs) to extract latent regional characteristics from satellite imagery, enabling effective revenue prediction despite small sample constraints. Our methodology integrates municipal-level demographic data with latent LMM-estimated features, capturing both explicit metrics and implicit characteristics that conventional methods overlook. Experiments across multiple clinic types and location conditions demonstrated that incorporating latent LMM-estimated features consistently improved prediction performance, with average improvements of 8.3% for mean squared error (MSE) and 31.6% for coefficient of determination (R \(^2\) ) compared to demographic-only approaches. The extracted interpretable features provide concrete decision-making criteria for practitioners. This versatile framework extends beyond healthcare to optimize location strategy across various industries facing similar data constraints.