Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating ChatGPT into Secure Hospital Networks: A Case Study on Improving Radiology Report Analysis

Kyungsu Kim, Junhyun Park, Saul Langarica, Adham Mahmoud Alkhadrawi, Synho Do

TL;DR

This study demonstrates the first in-hospital adaptation of a cloud-based AI, similar to ChatGPT, into a secure model for analyzing radiology reports, prioritizing patient data privacy and achieving over 95% accuracy in detecting anomalies.

Abstract

This study demonstrates the first in-hospital adaptation of a cloud-based AI, similar to ChatGPT, into a secure model for analyzing radiology reports, prioritizing patient data privacy. By employing a unique sentence-level knowledge distillation method through contrastive learning, we achieve over 95% accuracy in detecting anomalies. The model also accurately flags uncertainties in its predictions, enhancing its reliability and interpretability for physicians with certainty indicators. These advancements represent significant progress in developing secure and efficient AI tools for healthcare, suggesting a promising future for in-hospital AI applications with minimal supervision.

Integrating ChatGPT into Secure Hospital Networks: A Case Study on Improving Radiology Report Analysis

TL;DR

This study demonstrates the first in-hospital adaptation of a cloud-based AI, similar to ChatGPT, into a secure model for analyzing radiology reports, prioritizing patient data privacy and achieving over 95% accuracy in detecting anomalies.

Abstract

This study demonstrates the first in-hospital adaptation of a cloud-based AI, similar to ChatGPT, into a secure model for analyzing radiology reports, prioritizing patient data privacy. By employing a unique sentence-level knowledge distillation method through contrastive learning, we achieve over 95% accuracy in detecting anomalies. The model also accurately flags uncertainties in its predictions, enhancing its reliability and interpretability for physicians with certainty indicators. These advancements represent significant progress in developing secure and efficient AI tools for healthcare, suggesting a promising future for in-hospital AI applications with minimal supervision.
Paper Structure (23 sections, 4 equations, 7 figures, 5 tables)

This paper contains 23 sections, 4 equations, 7 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Baseline: Document-level Knowledge Distillation
  5. Label Extraction from Teacher Model
  6. Training and Inference for Student Model
  7. Proposed: Sentence-level Knowledge Distillation
  8. Label Extraction from Teacher Model
  9. Training and Inference for Student Model
  10. Objective Function for KD Training
  11. Results
  12. Setup
  13. Performance Comparison between Proposed and Baseline KD Approach
  14. Analysis of Potential Cause for S-KD Advancement
  15. Ablation Study for Contrastive Loss in KD Learning
  16. ...and 8 more sections

Figures (7)

  • Figure 1: Our study is the first to test the feasibility of distilling knowledge from cloud model like ChatGPT into a non-cloud model for radiology report analysis
  • Figure 2: Improving knowledge distillation performance and interpretability, our approach incorporates sentence-level knowledge distillation and enhances reliability by introducing an additional label (uncertain) for the network explicitly to indicate uncertainty in prediction results
  • Figure 3: Comparison of AD performance between S-KD and D-KD: S-KD demonstrates superior detection in abnormal (or normal) documents with fewer abnormal (or uncertain) sentences, outperforming D-KD in identifying challenging AD cases
  • Figure 4: Sample cases: Document-level vs Sentence-level KD - Demonstrating instances where document-level KD fails and sentence-level KD succeeds in accurately predicting abnormal and normal medical reports
  • Figure 5: Comparison of latent vector distribution for each class depending on whether the contrastive setup is used (ours) or not
  • ...and 2 more figures