A Comparative Study of Semi-supervised Deep Anomaly Detection and LLMs for Monitoring Patients with Severe Health Status Undergoing Radiotherapy

Yang Yan, Xinglei Shen, Ronald Chen, Hao Gao, Chen Zhao, Guihong Wan, Yi He, Zhong Chen

Keywords: Semi-supervised anomaly detection, Patient-reported outcomes (PROs), zero-shot Large Language Model, Llama 3.1-8B, Radiotherapy

TL;DR: This study assesses semi-supervised deep anomaly detection (AD) methods and novel zero-shot LLM prompts for identifying prostate cancer patients at risk of severe radiotherapy-induced symptoms via patient-reported outcomes (PROs).

Abstract: This study assesses semi-supervised deep anomaly detection (AD) methods and novel zero-shot LLM prompts for identifying prostate cancer patients at risk of severe radiotherapy-induced symptoms via patient-reported outcomes (PROs). While LLMs underperformed compared to semi-supervised AD models in key metrics (e.g., precision, recall, and F1-score), they provided valuable decision explanations and required no training data. This highlights their potential for straightforward clinical deployment without the need for extensive model development.

Submission Number: 15