ClinicalT5-PET-impression Open-source Model - Automatically Generate Personalized Clinical Impressions Based on PET Examinations

Clinicalt5 PET Impression

Developed by xtie

This is a fine-tuned version of the Clinical-T5 model for PET report findings summarization, capable of automatically generating personalized clinical impressions based on PET examination findings.

Large Language Model

Transformers

English#PET Report Summary #Personalized Impression Generation #Physician Style Adaptation

Downloads 17

Release Time : 9/3/2023

Model Overview

This model, fine-tuned from the Clinical-T5 model, is specifically designed for processing PET report data and can automatically generate personalized impression summaries that meet clinical needs based on examination findings.

Model Features

Personalized Impression Generation

The model can learn the reporting styles of different physicians and generate personalized clinical impressions that match specific physician styles.

High Clinical Utility

89% of the generated impressions were deemed clinically acceptable by physicians, with an average utility score of 4.08/5.

Domain-Specific Optimization

Specifically fine-tuned for PET report data, demonstrating excellent performance in the medical field.

Model Capabilities

Medical Text Summarization

Personalized Report Generation

Automatic Clinical Impression Generation

Use Cases

Medical Reports

Automatic PET Report Impression Generation

Automatically generates the clinical impression section based on PET examination findings

89% of the generated impressions were deemed clinically acceptable by physicians

Physician Style Adaptation

Generates clinical impressions that match the reporting style of specific physicians

Utility scores comparable to other physician-dictated impressions (4.03 vs 4.08)

🚀 Automatic Personalized Impression Generation for PET Reports Using Large Language Models 📄✍

This project utilizes large language models to automatically generate personalized impressions for PET reports, which is of great significance in the medical field, especially for improving the efficiency of PET report generation.

🚀 Quick Start

Model Usage

finetuned_model = "xtie/ClinicalT5-PET-impression"
tokenizer = AutoTokenizer.from_pretrained(finetuned_model) 
model = AutoModelForSeq2SeqLM.from_pretrained(finetuned_model, ignore_mismatched_sizes=True).eval()

findings_info =
"""
Description: PET CT WHOLE BODY
Radiologist: James
Findings:
Head/Neck: xxx Chest: xxx Abdomen/Pelvis: xxx Extremities/Musculoskeletal: xxx
Indication:
The patient is a 60-year old male with a history of xxx
"""

inputs = tokenizer(findings_info.replace('\n', ' '),
                  padding="max_length",
                  truncation=True,
                  max_length=1024,
                  return_tensors="pt")
input_ids = inputs.input_ids.to("cuda")
attention_mask = inputs.attention_mask.to("cuda")
outputs = model.generate(input_ids,
                        attention_mask=attention_mask,
                        max_new_tokens=512, 
                        num_beam_groups=1,
                        num_beams=4, 
                        do_sample=False,
                        diversity_penalty=0.0,
                        num_return_sequences=1, 
                        length_penalty=2.0,
                        no_repeat_ngram_size=3,
                        early_stopping=True
                        )
# get the generated impressions
output_str = tokenizer.decode(outputs,
                              skip_special_tokens=True)

✨ Features

The fine-tuned large language model provides clinically useful, personalized impressions based on PET findings.
To our knowledge, this is the first attempt to automate impression generation for whole-body PET reports.

📦 Installation

The README does not provide specific installation steps, so this section is skipped.

💻 Usage Examples

Basic Usage

finetuned_model = "xtie/ClinicalT5-PET-impression"
tokenizer = AutoTokenizer.from_pretrained(finetuned_model) 
model = AutoModelForSeq2SeqLM.from_pretrained(finetuned_model, ignore_mismatched_sizes=True).eval()

findings_info =
"""
Description: PET CT WHOLE BODY
Radiologist: James
Findings:
Head/Neck: xxx Chest: xxx Abdomen/Pelvis: xxx Extremities/Musculoskeletal: xxx
Indication:
The patient is a 60-year old male with a history of xxx
"""

inputs = tokenizer(findings_info.replace('\n', ' '),
                  padding="max_length",
                  truncation=True,
                  max_length=1024,
                  return_tensors="pt")
input_ids = inputs.input_ids.to("cuda")
attention_mask = inputs.attention_mask.to("cuda")
outputs = model.generate(input_ids,
                        attention_mask=attention_mask,
                        max_new_tokens=512, 
                        num_beam_groups=1,
                        num_beams=4, 
                        do_sample=False,
                        diversity_penalty=0.0,
                        num_return_sequences=1, 
                        length_penalty=2.0,
                        no_repeat_ngram_size=3,
                        early_stopping=True
                        )
# get the generated impressions
output_str = tokenizer.decode(outputs,
                              skip_special_tokens=True)

📚 Documentation

Model Description

This is the fine-tuned Clinical-T5 model for summarizing findings in PET reports.

To check our fine-tuned large language models (LLMs) for PET report summarization:

Abstract

Purpose: To determine if fine-tuned large language models (LLMs) can generate accurate, personalized impressions for whole-body PET reports.

Materials and Methods: Twelve language models were trained on a corpus of PET reports using the teacher-forcing algorithm, with the report findings as input and the clinical impressions as reference. An extra input token encodes the reading physician’s identity, allowing models to learn physician-specific reporting styles. Our corpus comprised 37,370 retrospective PET reports collected from our institution between 2010 and 2022. To identify the best LLM, 30 evaluation metrics were benchmarked against quality scores from two nuclear medicine (NM) physicians, with the most aligned metrics selecting the model for expert evaluation. In a subset of data, model-generated impressions and original clinical impressions were assessed by three NM physicians according to 6 quality dimensions and an overall utility score (5-point scale). Each physician reviewed 12 of their own reports and 12 reports from other physicians. Bootstrap resampling was used for statistical analysis.

Results: Of all evaluation metrics, domain-adapted BARTScore and PEGASUSScore showed the highest Spearman’s ρ correlations (0.568 and 0.563) with physician preferences. Based on these metrics, the fine-tuned PEGASUS model was selected as the top LLM. When physicians reviewed PEGASUS-generated impressions in their own style, 89% were considered clinically acceptable, with a mean utility score of 4.08/5. Physicians rated these personalized impressions as comparable in overall utility to the impressions dictated by other physicians (4.03, P=0.41).

Conclusion: Personalized impressions generated by PEGASUS were clinically useful, highlighting its potential to expedite PET reporting.

Read the full paper

Performance Metrics

For detailed evaluation results, please refer to our paper.

ROUGE-1: 53.2
ROUGE-2: 30.1
ROUGE-L: 39.4
BLEU: 23.9
BERTScore: 0.743

Hardware

The models were trained on NVIDIA A100 GPUs.

🔧 Technical Details

The README does not provide specific technical details, so this section is skipped.

📄 License

The README does not provide license information, so this section is skipped.

📁 Additional Resources

Finetuned from model: Clinical T5-Large
Codebase for training and inference: GitHub Repository

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご