Evaluation and progression analysis of pulmonary tuberculosis from digital chest radiographs¹

Abstract

Pulmonary tuberculosis is one of the most common infections in the third world countries. We propose a method for automated evaluation and progression analysis of this disease by processing a sequence of digital chest radiographs. The presence of radiological features characteristic to pulmonary tuberculosis is detected within the regions of interest using suitable image processing and pattern recognition techniques. The detected features are then quantified to evaluate the extent of the disease. The measures are evaluated for a sequence of radiographs of a patient and are normalized with respect to rib size measures. Temporal changes in these normalized measures indicate the progression of the disease during the period in which the sequence of radiographs is obtained. The details of the proposed technique and results of analysis of a number of tuberculous patients are given in the paper.

Introduction

Tuberculosis has been one of the most dreadful diseases in the Indian subcontinent. Poor living conditions, malnutrition and compromised immune response are the main reasons for the spread of this infection. The symptoms and the nature of the pathology in pulmonary tuberculosis have been described in details in Refs 1, 2. It may be mentioned here that certain pathological tests must be performed to confirm the diagnosis. The chest X-rays are analyzed by radiologists primarily to evaluate the level of infection. The details about various radiological features of tuberculosis have been discussed in Refs 3, 4. The treatment for this disease is usually carried out for a period ranging from 9 to 18$months with regular radiological examinations to check for the prognosis of the disease. During the period of treatment, the lesions heal and the abnormal features in the chest radiograph gradually reduce and finally only traces of fibrosed tissue remain. Our effort, in this paper, lies in building a vision-based system by which the recovery rate of the disease for each patient can be quantified [5]. This is done by processing a sequence of digital chest radiographs of the same patient and detecting tuberculous features in these images and subsequently quantifying them. Our effort does not envisage an early detection of the disease as it is pathologically not possible to do so from chest radiographs alone. The analysis and quantification of the recovery rate of a patient can have significant clinical relevance as it would indicate whether or not the patient is responding to the treatment at a proper rate.

A computer-aided diagnosis system based on some quantitative criteria and a knowledge of all possible abnormal features to aid diagnosis has been a subject of research in the past few decades 6, 7, 8, 9, 10, 11, 12, 13. It has already been reported by researchers [7]that for some special cases this improves the overall diagnostic accuracy of a radiologist. However, no work on the detection of radiological features related to pulmonary tuberculosis and their quantification has been reported in the literature to the best of our knowledge.

There have been several efforts 14, 15, 16, 17to detect the rib structure in the image so that this knowledge of the rib locations can be used for further processing of the image. In Ref. [6]the authors investigate the problem of segmentation as a pattern classification problem. Work on computer-aided diagnosis of diseases relating to the detection of interstitial infiltrates has been reported in Refs 7, 8, 9, the detection of pulmonary nodules in Refs 10, 11, and the change analysis of the heart shadow and the associated vasculature in Ref. [12]. Most of these efforts concentrate on characterizing a certain class of radiographic features and do not treat any particular disease as a whole. In Ref. [13]Krueger et al. considered the disease pneumoconiosis but there has not been any subsequent effort in quantifying the disease.

Among various radiological features [4]that manifest due to tuberculosis, the most common features are infiltration and cavitation. In this work we have considered only these two features and the detection and the quantification schemes for them are outlined. Other features, such as fibrosis and calcification, can be additionally used for the quantification of the healing process and will be incorporated subsequently. The major problem in working with digital chest radiographs is the superimposed rib shadows within the lung field. Filtering techniques cannot easily get around this problem without the loss of some of the important information. The filtering technique used in Ref. [18]was initially attempted to remove the rib structure, but it leads to a significant loss of information. The enhancement technique as described in Ref. [19]does improve the contrast but the rib shadows also get enhanced. In this work we have adopted a data-driven algorithmic approach, where the radiologist's diagnostic routine is modeled by a sequence of computer algorithms.

The organization of the paper is as follows. In the next section, the initial processing of the chest radiographs to locate different anatomical regions of interest is discussed. Specifically, the lung fields are located in the image and are divided into three zones of interest. The rib spaces are located and a rib-subtracted lung field region is obtained for the detection of features. Methods of detection of the infiltration regions and cavitations are discussed in Section 3and Section 4, respectively. An iterative segmentation [20]is used to detect the regions of infiltration and subsequently to locate cavities, if any. In Section 5, we extract the quantitative measures for these features and analyze them to learn the recovery rate or the progression of the disease over time. Relevant results are shown in every section that substantiate the suitability of the proposed technique in building a system for aiding automated diagnosis. Section 6concludes the paper along with a discussion on the future scope of the work.