Free-form Text Research Articles

P389 Comparative Evaluation of ChatGPT and Human Specialists in the Application of ECCO Guidelines for the Management of Inflammatory Bowel Diseases and Malignancies: A Proof-of-Concept Study

Abstract Background Societal guidelines on colorectal dysplasia screening, surveillance and endoscopic management in inflammatory bowel diseases (IBD) are rather complex, and physician adherence to them is suboptimal. We aimed to evaluate the use of ChatGPT, a large language model, in generating accurate guideline-based recommendations for colorectal dysplasia screening, surveillance and endoscopic management in IBD in line with European Crohn’s and Colitis Organization (ECCO) guidelines. Methods Thirty clinical scenarios in the form of free text regarding colorectal dysplasia in IBD were prepared and presented to ChatGPT and four gastroenterologists, two of them specializing in IBD and two with non-IBD specialties. Two additional IBD specialists subsequently assessed all responses provided by ChatGPT and the four gastroenterologists, judging their accuracy according to ECCO guidelines. Results ChatGPT provided accurate recommendations in 90% of cases (27/30), while among the four gastroenterologists the correct response rates were 28/30 (93%), 23/30 (77%), 26/30 (87%), and 25/30 (83%). The latter two represent the correct response rates of the IBD experts. No statistically significant differences were observed between the accuracy of ChatGPT versus all gastroenterologists (p=0.44), or between the accuracy of ChatGPT versus the IBD experts and non-IBD expert gastroenterologists (p=0.71). Conclusion This study highlights the potential of language models in enhancing guideline adherence regarding colorectal dysplasia in IBD. Further investigation of additional resources and prospective evaluation in real-world settings are warranted.

Investigating deep-learning NLP for automating the extraction of oncology efficacy endpoints from scientific literature

ObjectiveBenchmarking drug efficacy is a critical step in clinical trial design and planning. The challenge is that much of the data on efficacy endpoints is stored in scientific papers in free text form, so extraction of such data is currently a largely manual task. Our objective is to automate this task as much as possible. MethodsIn this study we have developed and optimised a framework to extract efficacy endpoints from text in scientific papers, using a machine learning approach. ResultsOur machine learning model predicts 25 classes associated with efficacy endpoints and leads to high F1 scores (harmonic mean of precision and recall) of 96.4 % on the test set, and 93.9 % and 93.7 % on two case studies. ConclusionThese methods were evaluated against – and showed strong agreement with – subject matter experts and show significant promise in the future of automating the extraction of clinical endpoints from free text. SignificanceClinical information extraction from text data is currently a laborious manual task which scales poorly and is prone to human error. Demonstrating the ability to extract efficacy endpoints automatically shows great promise for accelerating clinical trial design moving forwards.

Application of natural language processing to post-structuring of rectal cancer MRI reports

To evaluate a natural language processing (NLP) system for extracting structured information from the free-form text of rectal cancer magnetic resonance imaging (MRI) reports written in Chinese. A rule-based NLP model that could extract 11 key image features of rectal cancer was constructed using 358 MRI reports of rectal cancer written between 2015 and 2021. Fifty reports written before 2015 and 50 written after 2021 were used as test datasets, and the reference standard was determined by manual extraction of information by two radiologists. The length and reporting rate of image features in pre-2015 and post-2021 datasets, as well as the accuracy, precision, recall, and F1 score of feature extraction by the NLP system, were compared. The time required for the NLP to extract data was compared with that required by the radiologists. Reports written after 2021 had longer diagnostic impression sections than reports written before 2015. The reporting rate of key imaging features of rectal cancer was 36.55% before 2015 and 79.82% after 2021. The accuracy, precision, recall, and F1 score of NLP for correct extraction of values from reports were 93.82%, 95.63%, 87.06%, and 91.15%, respectively, for pre-2015 reports, and 92.55%, 98.53%, 94.15%, and 96.29%, respectively, for post-2021 reports. NLP generated all the structured information in <1 second. The NLP system with rule-based pattern matching achieved rapid and accurate structured processing of rectal cancer MRI reports. MRI reports with structured templates are more suitable for NLP-based extraction of information.

Extracting Clinical Information From Japanese Radiology Reports Using a 2-Stage Deep Learning Approach: Algorithm Development and Validation.

Radiology reports are usually written in a free-text format, which makes it challenging to reuse the reports. For secondary use, we developed a 2-stage deep learning system for extracting clinical information and converting it into a structured format. Our system mainly consists of 2 deep learning modules: entity extraction and relation extraction. For each module, state-of-the-art deep learning models were applied. We trained and evaluated the models using 1040 in-house Japanese computed tomography (CT) reports annotated by medical experts. We also evaluated the performance of the entire pipeline of our system. In addition, the ratio of annotated entities in the reports was measured to validate the coverage of the clinical information with our information model. The microaveraged F1-scores of our best-performing model for entity extraction and relation extraction were 96.1% and 97.4%, respectively. The microaveraged F1-score of the 2-stage system, which is a measure of the performance of the entire pipeline of our system, was 91.9%. Our system showed encouraging results for the conversion of free-text radiology reports into a structured format. The coverage of clinical information in the reports was 96.2% (6595/6853). Our 2-stage deep system can extract clinical information from chest and abdomen CT reports accurately and comprehensively.

Deep-learning-based natural-language-processing models to identify cardiovascular disease hospitalisations of patients with diabetes from routine visits’ text

Writing notes is the most widespread method to report clinical events. Therefore, most of the information about the disease history of a patient remains locked behind free-form text. Natural language processing (NLP) provides a solution to automatically transform free-form text into structured data. In the present work, electronic healthcare records data of patients with diabetes were used to develop deep-learning based NLP models to automatically identify, within free-form text describing routine visits, the occurrence of hospitalisations related to cardiovascular disease (CVDs), an outcome of diabetes. Four possible time windows of increasing level of expected difficulty were considered: infinite, 24 months, 12 months, and 6 months. Model performance was evaluated by means of the area under the precision recall curve, as well as precision, recall, and F1-score after thresholding. Results showed that the proposed NLP approach was successful for both the infinite and 24-month windows, while, as expected, performance deteriorated with shorter time windows. Possible clinical applications of tools based on the proposed NLP approach include the retrospective filling of medical records with respect to a patient’s CVD history for epidemiological and research purposes as well as for clinical decision making.

Police officers' perspective on doping and prevention among recreational athletes: a cross-sectional study.

The use of anabolic androgenic steroids among recreational athletes has received growing attention in recent decades. Several countries have implemented bans on doping; however, recreational athletes and other subpopulations continue to use doping substances. Recognizing that the police play a crucial role in preventing the use and dealing of doping substances in Sweden, efforts have been made to intensify police interventions and enhance collaboration with other key actors. This study examined police officers' perceptions of doping as defined in Swedish law, related problems, and suggestions for effective prevention of doping in the society. A cross-sectional survey study was conducted using a web survey of police officers (N = 597). Data were analyzed using descriptive statistics and free-form text responses were analyzed using content analysis. Participant responses to the survey (73.7% response rate) indicated that approximately 62.6% thought that doping is a societal problem, and approximately 26% perceived that the availability of doping substances has increased over the past three years. A total of 95.6% of respondents believed that doping occurred in connection with other crimes such as intimate partner violence (88.2%) and drug-related crimes (88.0%). Further, 96.3% of respondents perceived that it was their duty to prevent doping, but 63.8% indicated that doping-related work was not prioritized within their local police district. Police officers perceived doping as a societal problem and expressed motivation to counteract it, highlighting increased knowledge, legislative changes, intensified doping prevention in gyms, and commitment from other societal actors to increase the effectiveness of doping prevention. Suggestions for increasing the efficiency of doping prevention included education and increased knowledge at all levels in the police organization, intensified prevention efforts at gyms, legislative changes to permit simplified doping test procedures, and breach of secrecy for postal items. There was also a suggestion for further engagement from other actors, such as healthcare workers, school officials, and non-governmental organizations.

Using personal writings to detect dementia: A text mining approach.

A novel text mining pilot for dementia detection using Linguistic Inquiry and Word Count (LIWC) was tested on public figures' writings looking at word choice and affect compared to those with and without dementia. The differences found in this analysis mirror the expected patterns where writings of people with dementia reflect significantly more analytical thinking words, but significantly less authentic and emotional tone. In addition, the analysis found that people with dementia use significantly less functional words, such as grammar, and affections (happiness, sadness, anger, sadness), but tend to use significantly more pronouns in their writings. Written samples of those with dementia also use significantly less time-oriented words that indicate past, present, or future. The analysis of free form text suggests a potential avenue for detecting early changes that correlate with dementia, allowing for early preventative treatment before noticeable cognitive impairment occurs.

Unlocking new efficiencies: How structured content authoring is streamlining the production of clinical documents for the pharmaceutical industry

Current practice requires clinical and regulatory documents to be created and updated manually by medical writers throughout a product’s development. Conventionally, document content is unstructured, with freeform text, figures, and tables that the medical writer can arrange in any configuration. By structuring and standardising clinical and regulatory content, the pharmaceutical industry can shift from a document-based to a content-based approach. This transition will require adopting structured content management tools and common structures, and standardising content. In tandem, medical writers must evolve their skillset and ways of working, primarily through planning and producing content and adopting structured content authoring practices to facilitate content creation and reuse. This article introduces structured content authoring and outlines how the medical writing role in the pharmaceutical industry may soon evolve.

An investigation of municipal housing permit data for representation of the Canadian housing stock in building codes analysis

Development of energy codes requires an appropriate building stock representation. To this end, information available within municipal construction permit records (e.g., dwelling type, floor and footprint area, foundation type, number of above-grade storeys, availability of an attached garage, and number of bedrooms) represents an untapped opportunity. However, these records are often stored in a semi-structured fashion with relevant information scattered within free-form text description fields, which necessitates text analytics. This paper introduces a text analytics-based method, employing the association rule mining, for information retrieval from permit records to support building stock modelling process; and the method is demonstrated with permit records from small residential buildings in seven major Canadian cities comprising over 240,000 unique entries. The analysis highlights that detached housing is the most common housing type and the average dwelling floor area was between 174 and 266 m2. Each permit record is associated with existing Canadian housing archetypes. It is found that mid-to-large-size detached archetypes are more common than smaller row or semi-detached archetypes. A building performance simulation-based investigation revealed that assuming an equal weight for each archetype, instead of using the distributions identified from the permit records, causes an underestimation of the housing stock's energy use by 36%.

Discovering social determinants of health from case reports using natural language processing: algorithmic development and validation

BackgroundSocial determinants of health are non-medical factors that influence health outcomes (SDOH). There is a wealth of SDOH information available in electronic health records, clinical reports, and social media data, usually in free text format. Extracting key information from free text poses a significant challenge and necessitates the use of natural language processing (NLP) techniques to extract key information.ObjectiveThe objective of this research is to advance the automatic extraction of SDOH from clinical texts.Setting and dataThe case reports of COVID-19 patients from the published literature are curated to create a corpus. A portion of the data is annotated by experts to create ground truth labels, and semi-supervised learning method is used for corpus re-annotation.MethodsAn NLP framework is developed and tested to extract SDOH from the free texts. A two-way evaluation method is used to assess the quantity and quality of the methods.ResultsThe proposed NER implementation achieves an accuracy (F1-score) of 92.98% on our test set and generalizes well on benchmark data. A careful analysis of case examples demonstrates the superiority of the proposed approach in correctly classifying the named entities.ConclusionsNLP can be used to extract key information, such as SDOH factors from free texts. A more accurate understanding of SDOH is needed to further improve healthcare outcomes.

The diversity of volcanic hazard maps around the world: insights from map makers

The IAVCEI Working Group on Hazard Mapping has been active since 2014 and has facilitated several activities to enable sharing of experiences of how volcanic hazard maps are developed and used around the world. One key activity was a global survey of 90 map makers and practitioners to collect data about official, published volcanic hazard maps and how they were developed. The survey asked questions about map content, design, and input data, as well as about the map development process and key lessons learned. Here we present the results of this global survey, which are then used to quantitatively describe and summarise current practices in volcanic hazard map development.We received entries related to 89 volcanic hazard maps (78% long-term/background maps and 22% short-term/crisis hazard maps), covering a total of 80 volcanoes across 28 countries. Although most maps captured in the survey are volcano-scale maps of stratovolcanoes that show similar types of content, such as primary hazard footprints or zones, they vary greatly in input data, communication style, format, appearance, scale, content, and visual design. This diversity stems from a range of factors, including differences in map purpose, the methodology used, the level of understanding of past eruptive history, the prevailing scientific and cartographic practice at the time, the state of volcanic activity, and variations in culture, national map standards and legal requirements.Experiences and lessons shared by our respondents can be divided into six main themes: map design considerations; the process of map development; map audience and map user needs; hazard assessment approach; map availability and accessibility; and external (e.g., political) influences. Insights shared included the importance of: visual design elements, map testing and evaluation, working with stakeholders and end users to improve a map’s efficacy and relevance, and considering possible unanticipated uses of hazard maps. These free-form text insights (i.e., responses to open-ended questions) from map makers and practitioners familiar with the maps lend depth and clarity to our results. They provide a rich complement to our more quantitative analysis of design elements and of approaches used to determine and delineate map zones.Results from our global survey of hazard map makers and practitioners, together with insights from other key initiatives of the Working Group on Hazard Mapping such as the Volcanic Hazard Maps Database (VHMD; https://volcanichazardmaps.org/), provide a snapshot of the wide variety of volcanic hazard maps generated over the past decades, and improve our understanding of the diversity across volcanic hazard mapping practices. These initiatives represent important steps towards fulfilling the aims of the Working Group, namely, to construct a framework for a classification scheme for volcanic hazard maps and to promote harmonized terminology, as well as to identify and categorise good practices and considerations for volcanic hazard mapping.

Transitioning a project-based course between onsite and online. An experience report

We present an investigation regarding the challenges faced by student teams across four consecutive iterations of a team-focused, project-based course in software engineering. The studied period includes the switch to fully online activities in the spring of 2020, and covers the return to face to face teaching two years later. We cover the feedback provided by over 1500 students, collected in free-text form on the basis of a survey. A qualitative research method was utilized to discern and examine the challenges and perceived benefits of a course that was conducted entirely online. We show that technical challenges remain a constant in project-based courses, with time management being the most affected by the move to online. Students reported that the effective use of collaborative tools eased team organization and communication while online. We conclude by providing a number of action points regarding the integration of online activities in face-to-face course unfolding related to project management, communication tools, the importance of teamwork, and of active mentor participation.

Prediction of American Society of Anesthesiologists Physical Status Classification from preoperative clinical text narratives using natural language processing

BackgroundElectronic health records (EHR) contain large volumes of unstructured free-form text notes that richly describe a patient’s health and medical comorbidities. It is unclear if perioperative risk stratification can be performed directly from these notes without manual data extraction. We conduct a feasibility study using natural language processing (NLP) to predict the American Society of Anesthesiologists Physical Status Classification (ASA-PS) as a surrogate measure for perioperative risk. We explore prediction performance using four different model types and compare the use of different note sections versus the whole note. We use Shapley values to explain model predictions and analyze disagreement between model and human anesthesiologist predictions.MethodsSingle-center retrospective cohort analysis of EHR notes from patients undergoing procedures with anesthesia care spanning all procedural specialties during a 5 year period who were not assigned ASA VI and also had a preoperative evaluation note filed within 90 days prior to the procedure. NLP models were trained for each combination of 4 models and 8 text snippets from notes. Model performance was compared using area under the receiver operating characteristic curve (AUROC) and area under the precision recall curve (AUPRC). Shapley values were used to explain model predictions. Error analysis and model explanation using Shapley values was conducted for the best performing model.ResultsFinal dataset includes 38,566 patients undergoing 61,503 procedures with anesthesia care. Prevalence of ASA-PS was 8.81% for ASA I, 31.4% for ASA II, 43.25% for ASA III, and 16.54% for ASA IV-V. The best performing models were the BioClinicalBERT model on the truncated note task (macro-average AUROC 0.845) and the fastText model on the full note task (macro-average AUROC 0.865). Shapley values reveal human-interpretable model predictions. Error analysis reveals that some original ASA-PS assignments may be incorrect and the model is making a reasonable prediction in these cases.ConclusionsText classification models can accurately predict a patient’s illness severity using only free-form text descriptions of patients without any manual data extraction. They can be an additional patient safety tool in the perioperative setting and reduce manual chart review for medical billing. Shapley feature attributions produce explanations that logically support model predictions and are understandable to clinicians.

ProgPrompt: program generation for situated robot task planning using large language models

AbstractTask planning can require defining myriad domain knowledge about the world in which a robot needs to act. To ameliorate that effort, large language models (LLMs) can be used to score potential next actions during task planning, and even generate action sequences directly, given an instruction in natural language with no additional domain information. However, such methods either require enumerating all possible next steps for scoring, or generate free-form text that may contain actions not possible on a given robot in its current context. We present a programmatic LLM prompt structure that enables plan generation functional across situated environments, robot capabilities, and tasks. Our key insight is to prompt the LLM with program-like specifications of the available actions and objects in an environment, as well as with example that can be executed. We make concrete recommendations about prompt structure and generation constraints through ablation experiments, demonstrate state of the art success rates in VirtualHome household tasks, and deploy our method on a physical robot arm for tabletop tasks. Website and code at progprompt.github.io

Computing recommendations from free-form text

While searching for consumer goods, users frequently ask for suggestions from their peers by writing short free-form textual requests. For example, when searching for movies users may ask for “Drama movies with a mind-bending story and a surprise ending, such as Fight Club” in one of the many online discussion boards. Despite the recent developments in large language models (LLMs) and natural language processing (NLP), modern recommender systems still struggle to process such requests. Therefore, in this paper we evaluate several approaches for annotating structured information from such short, free-form natural language user texts to calculate recommendations. We set up this evaluation as a two phase processes including (a) identification of the best NLP approach to identify key elements of users’ requests, and (b) assessment of the quality of recommendations computed with such elements.For our evaluation, we use a gold-standard reddit movie recommendation dataset consisting of annotations, manually created by crowdworkers who extracted keywords, actor names and movie titles. Using this dataset we evaluate a collection of more than 30 NLP and five recommender approaches. In addition, we perform an ablation study to assess relative annotation importance for movie recommendations. We find that domain-specific deep learning models, trained on a subset of data as well as embedding-based recommendation approaches are able to match the recommendation performance of recommendations computed from manual annotations. These promising results warrant further investigation in automatic processing of short free-form texts for computation of recommendations. Specifically, we provide insights into which NLP models and configurations work best for automatically annotating free text to compute (movie) recommendations, hence substantially reducing the search space for combinations of NLP and recommendation algorithms in the movie and potentially other domains.

Natural language processing to convert unstructured COVID-19 chest-CT reports into structured reports

BackgroundStructured reporting has been demonstrated to increase report completeness and to reduce error rate, also enabling data mining of radiological reports. Still, structured reporting is perceived by radiologists as a fragmented reporting style, limiting their freedom of expression. PurposeA deep learning-based natural language processing method was developed to automatically convert unstructured COVID-19 chest CT reports into structured reports. MethodsTwo hundred-two COVID-19 chest CT were retrospectively reviewed by two experienced radiologists, who wrote for each exam a free-form text radiological report and coherently filled the template provided by the Italian Society of Medical and Interventional Radiology, used as ground-truth. A semi-supervised convolutional neural network was implemented to extract 62 categorical variables from the report. Two iterations were carried-out, the first without fine-tuning, the second one performing a fine-tuning. The performance was measured using the mean accuracy and the F1 mean score. An error analysis was performed to identify errors entirely attributable to incorrect processing of the model. ResultsThe algorithm achieved a mean accuracy of 93.7% and an F1 score 93.8% in the first iteration. Most of the errors were exclusively attributable to wrong inference (46%). In the second iteration the model achieved for both parameters 95,8% and percentage of errors attributable to wrong inference decreased to 26%. ConclusionsThe convolutional neural network achieved an optimal performance in the automated conversion of free-form text into structured radiological reports, overcoming all the limitation attributed to structured reporting and finally paving the way for data mining of radiological report.

RadioLOGIC, a healthcare model for processing electronic health records and decision-making in breast disease

Digital health data used in diagnostics, patient care, and oncology research continue to accumulate exponentially. Most medical information, and particularly radiology results, are stored in free-text format, and the potential of these data remains untapped. In this study, a radiological repomics-driven model incorporating medical token cognition (RadioLOGIC) is proposed to extract repomics (report omics) features from unstructured electronic health records and to assess human health and predict pathological outcome via transfer learning. The average accuracy and F1-weighted score for the extraction ofrepomics features using RadioLOGIC are 0.934 and 0.934, respectively, and 0.906 and 0.903 for the prediction of breast imaging-reporting and data system scores. The areas under the receiver operating characteristic curve forthe prediction of pathological outcome without and with transfer learning are0.912 and 0.945, respectively. RadioLOGIC outperforms cohort models in the capability to extract features and also reveals promise for checking clinical diagnoses directly from electronic health records.

Starvar: symptom-based tool for automatic ranking of variants using evidence from literature and genomes

BackgroundIdentifying variants associated with diseases is a challenging task in medical genetics research. Current studies that prioritize variants within individual genomes generally rely on known variants, evidence from literature and genomes, and patient symptoms and clinical signs. The functionalities of the existing tools, which rank variants based on given patient symptoms and clinical signs, are restricted to the coverage of ontologies such as the Human Phenotype Ontology (HPO). However, most clinicians do not limit themselves to HPO while describing patient symptoms/signs and their associated variants/genes. There is thus a need for an automated tool that can prioritize variants based on freely expressed patient symptoms and clinical signs.ResultsSTARVar is a Symptom-based Tool for Automatic Ranking of Variants using evidence from literature and genomes. STARVar uses patient symptoms and clinical signs, either linked to HPO or expressed in free text format. It returns a ranked list of variants based on a combined score from two classifiers utilizing evidence from genomics and literature. STARVar improves over related tools on a set of synthetic patients. In addition, we demonstrated its distinct contribution to the domain on another synthetic dataset covering publicly available clinical genotype–phenotype associations by using symptoms and clinical signs expressed in free text format.ConclusionsSTARVar stands as a unique and efficient tool that has the advantage of ranking variants with flexibly expressed patient symptoms in free-form text. Therefore, STARVar can be easily integrated into bioinformatics workflows designed to analyze disease-associated genomes.AvailabilitySTARVar is freely available from https://github.com/bio-ontology-research-group/STARVar.

Training medical students in motivational interviewing using a blended learning approach: a proof-of-concept study.

Difficulties in implementing behavior change in patients with chronic diseases are common in clinical practice. Motivational interviewing (MI) helps clinicians to support patients in overcoming ambivalence while maintaining self-determination. The inclusion of MI in German medical training curricula is still rare. Furthermore, the effects of systematic teaching of MI, especially via blended learning, have hardly been investigated. Medical students participated in three curricular events related to MI, consisting of instructional videos and theoretical and practical components in a blended learning format. The aim of the study was to investigate the effect of teaching MI in students' medical education. A controlled, non-randomized study was conducted with an intervention group and a control group. Both groups completed questionnaires on their experience and knowledge related to MI, completed a knowledge test and rated their satisfaction with the course. MI was taught in the 6th semester of medical coursework as part of a psychosomatic course, in the 8th semester during a psychiatry course and in the 9th semester during a weekly psychiatry clerkship. Data from the intervention group (n = 35) and control group (n = 14) were analyzed, with 65.7% of students participating in all three parts of the curriculum. Overall interest in learning MI was high, with M = 2.92 (SD = 1.00). The results indicate a greater increase in knowledge over time in the intervention group. The majority (62.86%) stated that the curriculum was relevant to their future career. Free-form text responses indicated a high level of satisfaction with practical relevance. This study demonstrates the usefulness of an MI curriculum for medical students. The integration of MI into medical curricula is a promising curricular addition to improve doctor-patient communication. Future research should address patient perceptions of MI competencies and the persistence of acquired competencies.

LFR-GAN: Local Feature Refinement based Generative Adversarial Network for Text-to-Image Generation

Text-to-image generation aims to generate images from text descriptions. Its main challenge lies in two aspects: (1) Semantic consistency, i.e., the generated images should be semantically consistent with the input text; and (2) Visual reality, i.e., the generated images should look like real images. To ensure text-image consistency, existing works mainly learn to establish the cross-modal representations via a text encoder and image encoder. However, due to the limited representation capability of the fixed-length embeddings and the flexibility of the free-form text descriptions, the learned text-to-image model is incapable of maintaining the semantic consistency between image local regions and fine-grained descriptions. As a result, the generated images sometimes miss some fine-grained attributes of the generated object, such as the color or shape of a part of the object. To address this issue, this paper proposes a Local Feature Refinement Based Generative Adversarial Network (LFR-GAN) , which first divides the text into some independent fine-grained attributes and generates an initial image, then refines the image details based on these attributes. The main contributions are three-fold: (1) An attribute modeling approach is proposed to model the fine-grained text descriptions by mapping them into representations of independent attributes, which provides more fine-grained details for image generation. (2) A local feature refinement approach is proposed to enable the generated image to form a complete reflection of the fine-grained attributes contained in the text description. (3) A multi-stage generation approach is proposed to realize the fine-grained manipulation of complex images progressively, which aims to improve the performance of the refinement and generate photo-realistic images. Extensive experiments on the CUB and Oxford102 datasets show the effectiveness of our LFR-GAN approach in both text-to-image generation and text-guided image manipulation tasks. Our LFR-GAN approach shows superior performance compared to the state-of-the-art methods. The codes will be released at https://github.com/PKU-ICST-MIPL/LFR-GAN_TOMM2023 .