Rosminah M. Kassim

Segmentation of the Lung Anatomy for High Resolution Computed Tomography (HRCT) Thorax Images

In diagnosing interstitial lung disease (ILD) using HRCT Thorax images, the radiologists required to view large volume of images (30 slices scanned at 10 mm interval or 300 slices scanned at 1 mm interval). However, in the development of scoring index to assess the severity of the disease, viewing 3 to 5 slices at the predetermined levels of the lung is suffice for the radiologist. To develop an algorithm to determine the severity of the ILD, it is important for the computer aided system to capture the main anatomy of the chest, namely the lung and heart at these 5 predetermined levels. In this paper, an automatic segmentation algorithm is proposed to obtain the shape of the heart and lung. In determine the quality of the segmentation, ground truth or manual tracing of the lung and heart boundary done by senior radiologist was compared with the result from the proposed automatic segmentation. This paper discussed five segmentation quality measurements that are used to measure the performance of the proposed segmentation algorithm, namely, the volume overlap error rate (VOE), relative volumetric agreement (RVA), average symmetric surface distance (ASSD), root mean square surface distance (RMSD) and Hausdorff distance (HD). The results showed that the proposed segmentation algorithm produced good quality segmentation for both right and left lung and may be used in the development of computer aided system application.

Texture-based classification for reticular pattern and ground glass opacity in high resolution computed tomography Thorax images

Lung disease is a global disease that affects a large group of people and is of growing interest to researches. The role of Computer Aided Diagnosis (CAD) systems to assist doctors to diagnose and detect disease is a beneficial one. Most techniques used to classify diseases stem from the textural method that is commonly associated with Gray Level Co-occurrence Matrix (GLCM). For this study, the objective is to classify the presence of two medical features in lung diseases which are Reticular Pattern (RP) and Ground Glass Opacity (GGO). A senior radiologist rates each slice and lung of a patient for the RP and GGO. Five slices of predetermined level of HRCT Thorax images that representing the whole lung of ten diseased patients and ten normal patients were used in this study. The textural features are extracted from each patient using the GLCM method. Classification was done using the WEKA, a machine learning tool. The classifiers used were the Naive Bayes, Multilayer Perceptron (MLP), K-Nearest Neighbor (KNN), Radial Basis Function Network, Random Forest and Decision Table classifiers. The classifiers showed that it is possible to have a classifier with an overall accuracy of 0.81 with RF classifier.

Comparing watershed and FCM segmentation in detecting reticular pattern for interstitial lung disease

Lung is an important organ in human respiratory system. However a group of lung diseases known as interstitial lung diseases (ILD) may affect the tissue and space around the air sacs of the lung that prohibit the transferring of enough oxygen into bloodstream. Presently, ILD patients are diagnosed manually by the medical practitioner based on the clinical findings and High-Resolution Computed Tomography (HRCT) thorax images. The process of diagnosing using HRCT images is time-consuming and the outcomes are subjective in nature. One of the indicators of the ILD is the existence of reticular pattern on the HRCT Thorax images. The severity of ILD basically depends on the coarseness of this reticular pattern. The research focuses on the segmentation of the reticular pattern on the infected region based on the grades given by the ILD scoring index; grade 0 - absent, grade 1 - fine intralobular fibrosis predominating, grade 2 - microcystic pattern with airspace less than 3mm in diameter, and grade 3 - larger cysts 3-6mm in diameter. This paper discussed the two segmentation techniques, watershed segmentation algorithm and Fuzzy C-Means (FCM). The study shows that both methods able to segment the reticular pattern for grade 2 and grade 3 of the disease. FCM yielded better result compared to the watershed in term of having higher accuracy of cyst detection and less over-segmented region.

Enhanced automatic lung segmentation using graph cut for Interstitial Lung Disease

Radiologists are known to suffer from fatigue and drop in diagnostic accuracy due to large number of slices to read and long working hours. A computer aided diagnosis (CAD) system could help lighten the workload. Segmentation is the first step in a CAD system. This study aims to propose an accurate automatic segmentation. This study deals with High Resolution Computed Tomography (HRCT) scans of the thorax for 15 healthy patients and 81 diseased lungs segregated to five levels based on anatomic landmarks by a senior radiologist. The method used in this study combines thresholding and normalized graph cut which is a combination of region and contour based methods. The way the graph cut is implemented with a rule of exclusion can offer some knowledge for greater accuracy of segmentation. The segmentation was compared to manual tracing done by a trained person who is familiar with lung images. The segmentation yielded 98.32% and 98.07% similarity for right lung (RL) and left lung (LL). The segmentation error of Relative Volume Difference (RVD) for both RL and LL are also low at 0.89% and -0.13% respectively. The Overlap Volume Errors (OVE) are low at 3.17% and 3.74% for RL and LL. Thus the automatic segmentation proposed was able to segment accurately across right and left lung and was able to segment severe diseased lungs.

Gabor texture measure in interstitial lung disease discrimination using high resolution computed tomography thorax images

High rsolution computed tomography (HRCT) thorax imaging is a trusted modality which is widely used in medical institutions to detect respiratory disease, in particular, interstitial lung disease (ILD) detection. Observing the enormous number of slices for single patient is time consuming and consequently may lead to human error. In order to overcome to this problem, developing computer aided diagnostic (CAD) systems will assist radiologists to make correct decision in shorter time when a single case is studied. A prototype CAD system using level 5 slice of HRCT images is proposed in this study with measuring textures derived from Gabor filtered images. Ability of proposed CAD system for pairwise discrimination of ILD, non-ILD and normal lungs is tested with a statistical data analysis procedure.

Regression as a Tool to Measure Segmentation Quality and Preliminary Indicator of Diseased Lungs

Segmentation of the lung from HRCT Thorax images was studied. An automatic method of determining segmentation area is proposed. High quality of segmentation is considered achieved when the segmented area from the proposed algorithm is almost identical to the area obtained from the manual tracings by lung expert ground truth. High correlation between the two types of segmented areas showed that regression may be used as a tool to measure segmentation quality. Supplementary information may also be obtained from the regression plot. Prediction interval may be used as a possible indicator of diseased whilst outliers may show or indicate low segmentation quality or a possible severity of the disease.

Performance evaluation of automated lung segmentation for High Resolution Computed Tomography (HRCT) thorax images

Segmentation is the preliminary steps in developing a computer aided diagnosis (CAD) system. Determining the quality of segmentation will be able to minimize errors in the CAD system. Ninety-six High Resolution Computed Tomography (HRCT) thorax images in DICOM format were obtained from the Department of Diag-nostic imaging of Kuala Lumpur, Malaysia consisting of Interstitial Lung Disease (ILD) cases, other lung related diseases (Non-ILD) cases and healthy (normal) cases. The study utilizes a framework of having five pre-determined levels of HRCT Thorax image slices based on lung anatomy selected by the radiologist. For the purpose of this study only Level 1 is used. The images were automatically segmented and compared with ground truth which the manual tracings done by a radiologist. Polyline distance metric and Euclidean distance were used to determine the quality of segmentation. The quality of the segmentation deteriorates when the polyline and Euclidean distance increases. Generally values above five pixels would yield poor segmentation quality. Using the Bland-Altman method and plot, it can be seen the level of agreement between polyline and Euclidean distance metrics as well as the quality of segmentation.

Performance evaluation system for automated lung segmentation system using HRCT Thorax images

Segmentation is an important process especially in a Computer Aided Diagnosis (CADx) system. There are various methods of segmentation. A large majority of these involve a threshold based approach. For this study full HRCT Thorax scans from 10 normal patients were analysed. This study proposes a performance evaluation system for segmentation system. The performance evaluation uses five measures which are Jaccards Index, Hausdorff Distance, Area Error, and Relative Volume Difference. A segmentation system previously developed for a limited five HRCT thorax segmentation system called ALSS was evaluated using the proposed system. The results from all performance measures showed that the segmentation system was consistent and was able to segment accurately across all the slices of the thorax.

Investigating solidity as a shape feature in the selection of HRCT thorax images

Scoring indices from high resolution computed tomography (HRCT) thorax images are essential for grading various changes in the abnormalities of the lung. The scoring index requires the radiologist to view and grade the signs and changes of the lung tissues at five predetermined levels of the HRCT images based on the lung anatomy; level 1:aortic arch, level 2: trachea carina, level 3: pulmonary hilar, level 4: pulmonary venous confluence and level 5: 1 to 2 cm above the dome of right hemidiaphragm. The enormous number of slices to be observed emphasize the need for a computer-aided system to assist in subsequent investigations. This paper propose a statistical shape feature, solidity of the heart, right lung and left lung, to be used in an automatic segmentation algorithm to retrieve level 4 and level 5. The segmentation algorithm used involved multilevel thresholding and multiscale morphological filtering. The quality of the segmentation was confirmed by the senior radiologist. No outlier was detected using the leave-one-out method (LOM) suggesting that the solidity shape feature is robust. The success rate of 82.35% for level 4 identification was achieved, whereas, the level 5 identification yielded 70.59% success rate which was obtained using the nearest-neighbour method. The proposed procedure suggests that the solidity shape feature of the heart, right lung and left lung may be used as one of the indicator for the discrimination and identification of level 4 and level 5 HRCT Thorax image.