Deep learning for classification of pediatric chest radiographs by WHO's standardized methodology

Background

The World Health Organization (WHO)-defined radiological pneumonia is a preferred endpoint in pneumococcal vaccine efficacy and effectiveness studies in children. Automating the WHO methodology may support more widespread application of this endpoint.

Methods

We trained a deep learning model to classify pneumonia CXRs in children using the World Health Organization (WHO)’s standardized methodology. The model was pretrained on CheXpert, a dataset containing 224,316 adult CXRs, and fine-tuned on PERCH, a pediatric dataset containing 4,172 CXRs. The model was then tested on two pediatric CXR datasets released by WHO. We also compared the model’s performance to that of radiologists and pediatricians.

Results

The average area under the receiver operating characteristic curve (AUC) for primary endpoint pneumonia (PEP) across 10-fold validation of PERCH images was 0.928; average AUC after testing on WHO images was 0.977. The model’s classification performance was better on test images with high inter-observer agreement; however, the model still outperformed human assessments in AUC and precision-recall spaces on low agreement images.

Conclusion

A deep learning model can classify pneumonia CXR images in children at a performance comparable to human readers. Our method lays a strong foundation for the potential inclusion of computer-aided readings of pediatric CXRs in vaccine trials and epidemiology studies.

Universal adversarial attacks on deep neural networks for medical image classification

BACKGROUND

Deep neural networks (DNNs) are widely investigated in medical image classification to achieve automated support for clinical diagnosis. It is necessary to evaluate the robustness of medical DNN tasks against adversarial attacks, as high-stake decision-making will be made based on the diagnosis. Several previous studies have considered simple adversarial attacks. However, the vulnerability of DNNs to more realistic and higher risk attacks, such as universal adversarial perturbation (UAP), which is a single perturbation that can induce DNN failure in most classification tasks has not been evaluated yet.

METHODS

We focus on three representative DNN-based medical image classification tasks (i.e., skin cancer, referable diabetic retinopathy, and pneumonia classifications) and investigate their vulnerability to the seven model architectures of UAPs.

RESULTS

We demonstrate that DNNs are vulnerable to both nontargeted UAPs, which cause a task failure resulting in an input being assigned an incorrect class, and to targeted UAPs, which cause the DNN to classify an input into a specific class. The almost imperceptible UAPs achieved > 80% success rates for nontargeted and targeted attacks. The vulnerability to UAPs depended very little on the model architecture. Moreover, we discovered that adversarial retraining, which is known to be an effective method for adversarial defenses, increased DNNs' robustness against UAPs in only very few cases.

CONCLUSION

Unlike previous assumptions, the results indicate that DNN-based clinical diagnosis is easier to deceive because of adversarial attacks. Adversaries can cause failed diagnoses at lower costs (e.g., without consideration of data distribution); moreover, they can affect the diagnosis. The effects of adversarial defenses may not be limited. Our findings emphasize that more careful consideration is required in developing DNNs for medical imaging and their practical applications.

Determination of disease severity in COVID-19 patients using deep learning in chest X-ray images

PURPOSE

Chest X-ray plays a key role in diagnosis and management of COVID-19 patients and imaging features associated with clinical elements may assist with the development or validation of automated image analysis tools. We aimed to identify associations between clinical and radiographic features as well as to assess the feasibility of deep learning applied to chest X-rays in the setting of an acute COVID-19 outbreak.

METHODS

A retrospective study of X-rays, clinical, and laboratory data was performed from 48 SARS-CoV-2 RT-PCR positive patients (age 60±17 years, 15 women) between February 22 and March 6, 2020 from a tertiary care hospital in Milan, Italy. Sixty-five chest X-rays were reviewed by two radiologists for alveolar and interstitial opacities and classified by severity on a scale from 0 to 3. Clinical factors (age, symptoms, comorbidities) were investigated for association with opacity severity and also with placement of central line or endotracheal tube. Deep learning models were then trained for two tasks: lung segmentation and opacity detection. Imaging characteristics were compared to clinical datapoints using the unpaired student's t-test or Mann-Whitney U test. Cohen's kappa analysis was used to evaluate the concordance of deep learning to conventional radiologist interpretation.

RESULTS

Fifty-six percent of patients presented with alveolar opacities, 73% had interstitial opacities, and 23% had normal X-rays. The presence of alveolar or interstitial opacities was statistically correlated with age (P = 0.008) and comorbidities (P = 0.005). The extent of alveolar or interstitial opacities on baseline X-ray was significantly associated with the presence of endotracheal tube (P = 0.0008 and P = 0.049) or central line (P = 0.003 and P = 0.007). In comparison to human interpretation, the deep learning model achieved a kappa concordance of 0.51 for alveolar opacities and 0.71 for interstitial opacities.

CONCLUSION

Chest X-ray analysis in an acute COVID-19 outbreak showed that the severity of opacities was associated with advanced age, comorbidities, as well as acuity of care. Artificial intelligence tools based upon deep learning of COVID-19 chest X-rays are feasible in the acute outbreak setting.

Comparing different deep learning architectures for classification of chest radiographs

Chest radiographs are among the most frequently acquired images in radiology and are often the subject of computer vision research. However, most of the models used to classify chest radiographs are derived from openly available deep neural networks, trained on large image datasets. These datasets differ from chest radiographs in that they are mostly color images and have substantially more labels. Therefore, very deep convolutional neural networks (CNN) designed for ImageNet and often representing more complex relationships, might not be required for the comparably simpler task of classifying medical image data. Sixteen different architectures of CNN were compared regarding the classification performance on two openly available datasets, the CheXpert and COVID-19 Image Data Collection. Areas under the receiver operating characteristics curves (AUROC) between 0.83 and 0.89 could be achieved on the CheXpert dataset. On the COVID-19 Image Data Collection, all models showed an excellent ability to detect COVID-19 and non-COVID pneumonia with AUROC values between 0.983 and 0.998. It could be observed, that more shallow networks may achieve results comparable to their deeper and more complex counterparts with shorter training times, enabling classification performances on medical image data close to the state-of-the-art methods even when using limited hardware.

Label Co-Occurrence Learning With Graph Convolutional Networks for Multi-Label Chest X-Ray Image Classification

Existing multi-label medical image learning tasks generally contain rich relationship information among pathologies such as label co-occurrence and interdependency, which is of great importance for assisting in clinical diagnosis and can be represented as the graph-structured data. However, most state-of-the-art works only focus on regression from the input to the binary labels, failing to make full use of such valuable graph-structured information due to the complexity of graph data. In this paper, we propose a novel label co-occurrence learning framework based on Graph Convolution Networks (GCNs) to explicitly explore the dependencies between pathologies for the multi-label chest X-ray (CXR) image classification task, which we term the "CheXGCN". Specifically, the proposed CheXGCN consists of two modules, i.e., the image feature embedding (IFE) module and label co-occurrence learning (LCL) module. Thanks to the LCL model, the relationship between pathologies is generalized into a set of classifier scores by introducing the word embedding of pathologies and multi-layer graph information propagation. During end-to-end training, it can be flexibly integrated into the IFE module and then adaptively recalibrate multi-label outputs with these scores. Extensive experiments on the ChestX-Ray14 and CheXpert datasets have demonstrated the effectiveness of CheXGCN as compared with the state-of-the-art baselines.

A hybrid knowledge-guided detection technique for screening of infectious pulmonary tuberculosis from chest radiographs

Tuberculosis (TB) is a deadly infectious disease and the presence of cavities in the upper lung zones is a strong indicator that the disease has developed into a highly infectious state. Currently, the detection of TB cavities is mainly conducted by clinicians observing chest radiographs. Diagnoses performed by radiologists are labor intensive and very often there is insufficient healthcare personnel available, especially in remote communities. After assessing existing approaches, we propose an automated segmentation technique which takes a hybrid knowledge-based Bayesian classification approach to detect TB cavities automatically. We apply gradient inverse coefficient of variation (GICOV) and circularity measures to classify detected features and confirm true TB cavities. By comparing with non hybrid approaches and the classical active contour techniques for feature extraction in medical images, experimental results demonstrate that our approach achieves high accuracy with a low false positive rate in detecting TB cavities.

Indexing thoracic CT reports using a preliminary version of a standardized radiological lexicon (RadLex)

INTRODUCTION

To validate a preliminary version of a radiological lexicon (RadLex) against terms found in thoracic CT reports and to index report content in RadLex term categories.

MATERIAL AND METHODS

Terms from a random sample of 200 thoracic CT reports were extracted using a text processor and matched against RadLex. Report content was manually indexed by two radiologists in consensus in term categories of Anatomic Location, Finding, Modifier, Relationship, Image Quality, and Uncertainty. Descriptive statistics were used and differences between age groups and report types were tested for significance using Kruskal-Wallis and Mann-Whitney Test (significance level <0.05).

RESULTS

From 363 terms extracted, 304 (84%) were found and 59 (16%) were not found in RadLex. Report indexing showed a mean of 16.2 encoded items per report and 3.2 Finding per report. Term categories most frequently encoded were Modifier (1,030 of 3,244, 31.8%), Anatomic Location (813, 25.1%), Relationship (702, 21.6%) and Finding (638, 19.7%). Frequency of indexed items per report was higher in older age groups, but no significant difference was found between first study and follow up study reports. Frequency of distinct findings per report increased with patient age (p < 0.05).

CONCLUSION

RadLex already covers most terms present in thoracic CT reports based on a small sample analysis from one institution. Applications for report encoding need to be developed to validate the lexicon against a larger sample of reports and address the issue of automatic relationship encoding.

An ontology for PACS integration

An ontology describes a set of classes and the relationships among them. We explored the use of an ontology to integrate picture archiving and communication systems (PACS) with other information systems in the clinical enterprise. We created an ontological model of thoracic radiology that contained knowledge of anatomy, imaging procedures, and performed procedure steps. We explored the use of the model in two use cases: (1) to determine examination completeness and (2) to identify reference (comparison) images obtained in the same imaging projection. The model incorporated a total of 138 classes, including radiology orderables, procedures, procedure steps, imaging modalities, patient positions, and imaging planes. Radiological knowledge was encoded as relationships among these classes. The ontology successfully met the information requirements of the two use-case scenarios. Ontologies can represent radiological and clinical knowledge to integrate PACS with the clinical enterprise and to support the radiology interpretation process.

Projection profile analysis for identifying different views of chest radiographs

RATIONALE AND OBJECTIVES

For computerized analysis of chest images in the clinical environment, identification of frontal (posteroanterior/anteroposterior) and lateral chest radiographs is an important preprocessing step. In this study, we developed a method to distinguish frontal from lateral views of the chest radiographs based on an analysis of the projection profile.

MATERIALS AND METHODS

Projection profile is obtained by projecting a chest image on to the mediolateral axis. Two indices, body symmetry index and background percentage index, are computed from the projection profile. The combination of body symmetry index and background percentage index is used to determine the view of chest radiographs. The method is evaluated on a sample of 2000 frontal and 1000 lateral chest images.

RESULTS

The values of body symmetry index are found to be 1.18 +/- 0.23 and 3.07 +/- 1.42 for frontal and lateral chest images, respectively. The values of background percentage index are found to be 0.03 +/- 0.05 and 0.33 +/- 0.09 for frontal and lateral chest images, respectively. The discrimination is evaluated by linear discriminant analysis and receiver operating characteristic analysis. Area Az under the receiver operating characteristic curve with the combination of the two indices is 0.993.

CONCLUSION

The method can be used as a preprocessing step for further analysis in chest radiographs.

A natural language processing system to extract and code concepts relating to congestive heart failure from chest radiology reports

We have developed a natural language processing system for extracting and coding clinical data from free text reports. The system is designed to be easily modified and adapted to a variety of free text clinical reports such as admission notes, radiology and pathology reports, and discharge summaries. This report presents the results of this system to extract and code clinical concepts related to congestive heart failure from 39,000 chest radiology reports. The system detects the presence or absence of six concepts: congestive heart failure, Kerley B lines, cardiomegaly, prominent pulmonary vasculature, pulmonary edema, and pleural effusion. We compared it's output to a gold standard which consisted of specially trained human coders as well as an experienced physician. Results indicate that the system had high specificity, recall and precision for each of the concepts it is designed to detect.

Comparison of "B" readers' interpretations of chest radiographs for asbestos related changes

RATIONALE AND OBJECTIVES

The purpose of this study was to determine if chest radiographic interpretations by physicians retained by attorneys representing persons alleging respiratory changes from occupational exposure to asbestos would be confirmed by independent consultant readers.

MATERIALS AND METHODS

For 551 chest radiographs read as positive for lung changes by initial "B" readers retained by plaintiffs' attorneys, 492 matching interpretative reports were made available to the authors. Six consultants in chest radiology, also B readers, agreed to re-interpret the radiographs independently without knowledge of their provenance. The film source, patient name, and other identifiers on each film were masked. The International Labor Office 1980 Classification of Chest Radiographs(ILO 80) was used with forms designed by the US National Institute of Occupational Safety and Health to record the consult-ants' findings. The results were compared with initial readings for film quality, complete negativity, parenchymal abnormalities,small opacities profusion, and pleural abnormalities using chi-square tests and kappa statistics.Results. Initial readers interpreted study radiographs as positive for parenchymal abnormalities (ILO small opacity profusion category of 1/0 or higher) in 95.9% of 492 cases. Six consultants classified the films as 1/0 or higher in 4.5% of 2,952 readings. Statistical tests of these and other comparable data from the study showed highly significant differences between the interpretations of the initial readers and the findings of the consultants.

CONCLUSION

The magnitude of the differences between the interpretations by initial readers and the six consultants is too great to be attributed to interobserver variability. There is no support in the literature on x-ray studies of workers exposed to asbestos and other mineral dusts for the high level of positive findings recorded by the initial readers in this report.

Creating a text classifier to detect radiology reports describing mediastinal findings associated with inhalational anthrax and other disorders

OBJECTIVE

The aim of this study was to create a classifier for automatic detection of chest radiograph reports consistent with the mediastinal findings of inhalational anthrax.

DESIGN

The authors used the Identify Patient Sets (IPS) system to create a key word classifier for detecting reports describing mediastinal findings consistent with anthrax and compared their performances on a test set of 79,032 chest radiograph reports.

MEASUREMENTS

Area under the ROC curve was the main outcome measure of the IPS classifier. Sensitivity and specificity of an initial IPS model were calculated based on an existing key word search and were compared against a Boolean version of the IPS classifier.

RESULTS

The IPS classifier received an area under the ROC curve of 0.677 (90% CI = 0.628 to 0.772) with a specificity of 0.99 and maximum sensitivity of 0.35. The initial IPS model attained a specificity of 1.0 and a sensitivity of 0.04.

CONCLUSION

The IPS system is a useful tool for helping domain experts create a statistical key word classifier for textual reports that is a potentially useful component in surveillance of radiographic findings suspicious for anthrax.

Automated extraction and normalization of findings from cancer-related free-text radiology reports

We describe the performance of a particular natural language processing system that uses knowledge vectors to extract findings from radiology reports. LifeCode (A-Life Medical, Inc.) has been successfully coding reports for billing purposes for several years. In this study, we describe the use of LifeCode to code all findings within a set of 500 cancer-related radiology reports against a test set in which all findings were manually tagged. The system was trained with 1400 reports prior to running the test set.

RESULTS

LifeCode had a recall of 84.5% and precision of 95.7% in the coding of cancer-related radiology report findings.

CONCLUSION

Despite the use of a modest sized training set and minimal training iterations, when applied to cancer-related reports the system achieved recall and precision measures comparable to other reputable natural language processors in this domain.

The role of domain knowledge in automating medical text report classification

OBJECTIVE

To analyze the effect of expert knowledge on the inductive learning process in creating classifiers for medical text reports.

DESIGN

The authors converted medical text reports to a structured form through natural language processing. They then inductively created classifiers for medical text reports using varying degrees and types of expert knowledge and different inductive learning algorithms. The authors measured performance of the different classifiers as well as the costs to induce classifiers and acquire expert knowledge.

MEASUREMENTS

The measurements used were classifier performance, training-set size efficiency, and classifier creation cost.

RESULTS

Expert knowledge was shown to be the most significant factor affecting inductive learning performance, outweighing differences in learning algorithms. The use of expert knowledge can affect comparisons between learning algorithms. This expert knowledge may be obtained and represented separately as knowledge about the clinical task or about the data representation used. The benefit of the expert knowledge is more than that of inductive learning itself, with less cost to obtain.

CONCLUSION

For medical text report classification, expert knowledge acquisition is more significant to performance and more cost-effective to obtain than knowledge discovery. Building classifiers should therefore focus more on acquiring knowledge from experts than trying to learn this knowledge inductively.

Development of a template model to represent the information content of chest radiology reports

The authors describe the application of a methodology for developing representational models for loosely structured medical domains. The methodology is subdivided in two interrelated tasks: terminology acquisition and template generation. The methodology is applied to the domain of chest radiology, producing a domain-specific lexicon and a series of templates to represent all the relevant clinical information stated on a chest x-ray report. Details about the successive application of the methodology and the resulting versions of the lexicon and templates are presented. The most relevant aspects of the methodology utilization are discussed and compared with evidence from other authors.

Computer-aided diagnosis in chest radiography: a survey

The traditional chest radiograph is still ubiquitous in clinical practice, and will likely remain so for quite some time. Yet, its interpretation is notoriously difficult. This explains the continued interest in computer-aided diagnosis for chest radiography. The purpose of this survey is to categorize and briefly review the literature on computer analysis of chest images, which comprises over 150 papers published in the last 30 years. Remaining challenges are indicated and some directions for future research are given.

Medical text representations for inductive learning

Inductive learning algorithms have been proposed as methods for classifying medical text reports. Many of these proposed techniques differ in the way the text is represented for use by the learning algorithms. Slight differences can occur between representations that may be chosen arbitrarily, but such differences can significantly affect classification algorithm performance. We examined 8 different data representation techniques used for medical text, and evaluated their use with standard machine learning algorithms. We measured the loss of classification-relevant information due to each representation. Representations that captured status information explicitly resulted in significantly better performance. Algorithm performance was dependent on subtle differences in data representation.

Evaluation of SNOMED3.5 in representing concepts in chest radiology reports: integration of a SNOMED mapper with a radiology reporting workstation

Standardized medical terminologies are gaining importance in the representation of medical data. In this paper, we present the evaluation of the SNOMED3.5 medical terminology to code concepts routinely used in chest radiology reports. Integration of this terminology mapper into a radiology reporting workstation that incorporates a speech recognition system and a natural language processor is also discussed. A total of 700 anatomical location terms (including synonyms) were tested and 72% of the terms had corresponding SNOMED terms. Of the 28% that did not result in a match, 16% were either morphological variants of SNOMED terms or could be found from a combination of terms from two or more SNOMED axes. Only 12% of the terms (primarily specialized radiology terms) were concepts not actually included in the SNOMED terminology.

A visualization of cross-over data using linear functions

Previous work has shown that cross-over trial data can be analysed using within-subject linear functions. Scores that result from linear functions are graphed in quantile comparison plots in order to visualize the differences between factor levels. An example suggests how this visualization can be used to identify outliers or to provide a more specific interpretation of results. Additional examples indicate how this approach can be used to track carry-over differences or interaction effects. This article is a US Government work and is in the public domain in the United States.

Automatic identification of pneumonia related concepts on chest x-ray reports

A medical language processing system called SymText, two other automated methods, and a lay person were compared against an internal medicine resident for their ability to identify pneumonia related concepts on chest x-ray reports. Sensitivity (recall), specificity, and positive predictive value (precision) are reported with respect to an independent panel of physicians. Overall the performance of SymText was similar to the physician and superior to the other methods. The automatic encoding of pneumonia concepts will support clinical research, decision making, computerized clinical protocols, and quality assurance in a radiology department.

Comparing expert systems for identifying chest x-ray reports that support pneumonia

We compare the performance of four computerized methods in identifying chest x-ray reports that support acute bacterial pneumonia. Two of the computerized techniques are constructed from expert knowledge, and two learn rules and structure from data. The two machine learning systems perform as well as the expert constructed systems. All of the computerized techniques perform better than a baseline keyword search and a lay person, and perform as well as a physician. We conclude that machine learning can be used to identify chest x-ray reports that support pneumonia.

Selection of subtle cases for observer-performance studies: the importance of knowing the true diagnosis

RATIONALE AND OBJECTIVES

To assess the usefulness of classifying degree of difficulty in abnormality detection and to determine the effect of knowing the true diagnosis when selecting subtle images for observer-performance studies.

MATERIALS AND METHODS

A total of 529 posteroanterior chest images that had been used in a multiabnormality, multireader observer-performance study were rated by three observers as to the difficulty of determining the presence or absence of each abnormality when the true diagnosis was known and when it was not known. Changes in image subtlety ratings were evaluated, and actual observer-performance results for the different groups of images grouped according to raters' classifications with and without availability of the true diagnosis were compared.

RESULTS

The majority of negative cases (9,168 of 12,258, 74.8%) were rated as "easy" to determine. Substantial changes were made during the selection of the "subtle" case category when the truth was known compared with when the truth was not provided. These changes caused differences between typical and subtle cases in terms of observer performance. Combined ratings of case subtlety by agreement of multiple classifiers resulted in a well-ordered selection with decreasing observer performance as a function of subtlety ratings.

CONCLUSION

Cases for observer-performance studies that stress the diagnostic system can be successfully selected in the multiple-disease setting by experienced readers and should be selected with the truth known to the raters. The degree of agreement by multiple raters can be used to refine subtlety ratings.

Towards a comprehensive medical language processing system: methods and issues

Natural language processing (NLP) systems can help solve the data entry problem by providing coded data from textual reports for clinical applications. A number of NLP systems have shown promise, but have not yet achieved wide-spread use for practical applications. In order to achieve such use, a system must have broad coverage of the clinical domain and not be restricted to limited applications. In addition, an NLP system must perform satisfactorily for real-world applications. This paper describes methods and issues associated with an ongoing extension of MedLEE, an operational NLP system, from a limited domain to a domain that encompasses comprehensive clinical information.

Identification of suspected tuberculosis patients based on natural language processing of chest radiograph reports

Identification of eligible patients from electronically available patient data is a key difficulty in computerizing clinical practice guidelines because a large amount of the relevant data is stored as free text. We have been using MedLEE (Medical Language Extraction and Encoding System), a natural language processing system, to encode the clinical information in all chest radiograph and mammogram reports. This paper describes a retrospective study to determine if MedLEE can identify patients at risk for having tuberculosis (TB) based on their admission chest radiographs. Reports of 171 adult inpatients with culture-positive TB during 1992 and 1993 were manually coded (by a TB specialist) using seven terms suggestive of TB, and were also encoded by MedLEE. Using manual coding as the gold standard, MedLEE agreed on the classification of 152/171 (88.9%) reports--129/142 (90.8%) suspicious for TB and 23/29 (79.3%) not suspicious for TB; and 1072/1197 (89.6%) terms indicative of TB. Analysis showed that most of the discrepancies were caused by MedLEE not finding the location of the infiltrate. By ignoring the location of the infiltrate, the agreement became 157/171 (91.8%) reports and 946/1026 (92.2%) terms. Thus, natural language processing offers a practical alternative for using free-text reports to determine patient eligibility for computerized clinical practice guidelines.

Computerized analysis of interstitial infiltrates on chest radiographs: a new scheme based on geometric pattern features and Fourier analysis

RATIONALE AND OBJECTIVES

Detection of interstitial infiltrates on chest radiographs is difficult and subjective. Therefore, we developed a computerized method to provide quantitative analysis of lung texture to increase diagnostic accuracy.

METHODS

Two hundred chest radiographs--100 healthy and 100 abnormal with interstitial infiltrates--were digitized using a laser scanner. They were analyzed by an automated computerized scheme that uses a combination of two methods for detection of interstitial infiltrates: a lung texture analysis based on the Fourier transform and a geometric pattern feature analysis based on filtering techniques.

RESULTS

The overall sensitivity and specificity of the computerized scheme were 92% and 90%, respectively. The scheme achieved a sensitivity of 80% in subtle cases (n = 15) and 88% in cases with localized interstitial disease (n = 26), whereas the specificity remained unchanged. There was good correlation between the computer output and the radiologists' severity rating.

CONCLUSION

This enhanced computerized scheme exhibits high sensitivity and specificity with a large database.

An empirical investigation into the conceptual structure of chest radiograph findings

A method of investigating the conceptual structure of findings is presented, in which: 1) finding statements are extracted manually from free text, 2) the main concepts in each finding are manually identified and classified, and 3) the resulting sets of classes are examined for insights into finding structure. This study applies the method to chest radiograph reports. Although the subjects of the findings studied fall into seven classes, the same conceptual structure can be used for most of them.

Epidemiological data on US coal miners' pneumoconiosis, 1960 to 1988

OBJECTIVES

Statistics on prevalence of pneumoconiosis among working underground coal miners based on epidemiologic data collected between 1960 and 1988 are presented. The main intent was to examine the time-related trend in prevalence, particularly after 1969, when substantially lower dust levels were mandated by federal act.

METHODS

Data from studies undertaken between 1960 and 1968 were collected and compared. Information for the period 1969 to 1988 was extracted from a large ongoing national epidemiologic study. Tenure-specific prevalence rates and summary statistics derived from the latter data for four consecutive time intervals within the 19-year period were calculated and compared.

RESULTS

The results indicate a reduction in pneumoconiosis over time. The trend is similar to that seen in a large radiologic surveillance program of underground miners operated concurrently.

CONCLUSIONS

Although such factors as x-ray reader variation, changes in x-ray standards, and worker self-selection for examination may have influenced the findings to some extent, adjusted summary rates reveal a reduction in prevalence concurrent with reductions in coal mine dust levels mandated by federal act in 1969.

Surveillance data on US coal miners' pneumoconiosis, 1970 to 1986

OBJECTIVES

Statistics on prevalence of pneumoconiosis among working underground coal miners from data collected as part of a large national radiographic surveillance program between 1970 and 1986 are presented. The main intent was to examine the time-related trend in prevalence over this period, which coincides with historically low dust levels mandated by federal act.

METHODS

Tenure-specific prevalence rates and summary statistics derived from them for four consecutive time intervals within the 16-year period were calculated and compared.

RESULTS

The results indicate a reduction in pneumoconiosis over the life of the program. This trend is similar to that seen in epidemiologic studies undertaken concurrently.

CONCLUSIONS

Although low participation in the surveillance program and other problems complicate the findings, it appears that reductions in dust exposure mandated by federal act in 1969 have led to lower prevalence of pneumoconiosis among underground coal miners.

[A study of chest X-ray classification of farmer's lung--correlation between X-ray pattern and background factors]

Patients with farmer's lung were classified into five groups according to their chest X-ray pattern, and the background of each group was studied with respect to stage, severity, and prognosis. Type I group, which has fine nodular shadows on chest X-ray, showed a significantly high percentage of lymphocytes in BAL cells (p less than 0.01). Type V group, which has irregular patchy shadows on chest X-ray, showed increased peripheral WBC counts, increased CRP score, and decreased PaO2 compared to the other groups. Type II, III, and IV groups, which are characterized by combined X-ray patterns such as nodular shadows, ring-like shadows, and irregular pulmonary markings, had no specific background features according to our study. Our results suggest that type I and type V groups are likely to represent acute phase and severe phase, respectively. We could not identify any significant factors indicative of chronic disease in terms of restrictive and diffusion impairment of pulmonary function tests and pathologic findings obtained by TBLB. The chest X-ray classification we proposed is useful for the diagnosis of the acute and severe stages of farmer's lung.

Classification of chest radiographs for epidemiological purposes by people not experienced in the radiology of pneumoconiosis

Under controlled conditions 16 people (eight non-medical) inexperienced in the radiology of occupational lung diseases repeatedly classified 300 selected chest radiographs using the 1971 ILO U/C International Classification of Radiographs of Pneumoconioses. Eight experienced medical readers had previously classified 220 of the selected radiographs for profusion of small rounded opacities. Variability among readers was greater in experimental panels than among the experienced readers. But the average consistency between pairs of novice readers in their use of the 12 categories of profusion for the same radiographs was similar (about 29%) to the average consistency among the experienced readers. Subsequent work with nine of the participants showed that eight of them were able to produce classifications of coal miners' chest radiographs that correlated well with estimates the miners' exposures to respirable coal mine dust. It is concluded that the ILO classification scheme provides a sound descriptive system for recording the appearances of chest radiographs. Under controlled conditions the scheme may be used for epidemiological studies by those with no specialist knowledge or clinical experience. This presupposes that the radiographs concerned will have been examined previously for diagnostic purposes by a suitable qualified physician.