New approach to detect and classify stroke in skull CT images via analysis of brain tissue densities

Highlights

• A new algorithm (ABTD) is proposed to extract image features based on human brain tissue densities in medical CT images.

• ABTD is used to extract features from brain CT images.

• ABTD is compared against to five traditional feature extraction methods.

• Influence of the extraction method on the classification accuracy assessed using five machine learning techniques.

• Results confirm the superiority and suitability for medical images of ABTD.

Abstract

Background and Objective

Cerebral vascular accident (CVA), also known as stroke, is an important health problem worldwide and it affects 16 million people worldwide every year. About 30% of those that have a stroke die and 40% remain with serious physical limitations. However, recovery in the damaged region is possible if treatment is performed immediately. In the case of a stroke, Computed Tomography (CT) is the most appropriate technique to confirm the occurrence and to investigate its extent and severity. Stroke is an emergency problem for which early identification and measures are difficult; however, computer-aided diagnoses (CAD) can play an important role in obtaining information imperceptible to the human eye. Thus, this work proposes a new method for extracting features based on radiological density patterns of the brain, called Analysis of Brain Tissue Density (ABTD).

Methods

The proposed method is a specific approach applied to CT images to identify and classify the occurrence of stroke diseases. The evaluation of the results of the ABTD extractor proposed in this paper were compared with extractors already established in the literature, such as features from Gray-Level Co-Occurrence Matrix (GLCM), Local binary patterns (LBP), Central Moments (CM), Statistical Moments (SM), Hu’s Moment (HM) and Zernike’s Moments (ZM). Using a database of 420 CT images of the skull, each extractor was applied with the classifiers such as MLP, SVM, kNN, OPF and Bayesian to classify if a CT image represented a healthy brain or one with an ischemic or hemorrhagic stroke.

Results

ABTD had the shortest extraction time and the highest average accuracy (99.30%) when combined with OPF using the Euclidean distance. Also, the average accuracy values for all classifiers were higher than 95%.

Conclusions

The relevance of the results demonstrated that the ABTD method is a useful algorithm to extract features that can potentially be integrated with CAD systems to assist in stroke diagnosis.

Introduction

Stroke is an injury that abruptly affects brain tissues. This disease is caused by a change in the blood supply to a particular region of the brain, and it results in the loss or reduction of its related functions.

Cerebral vascular accidents (CVA) affect 16 million people worldwide every year, and 6 million of these people die [1]. However, another important problem related to strokes, besides mortality, is that many survivors have chronic consequences that are complex and heterogeneous. Studies estimate that there are about 2 million people with sequels to strokes in the United States. Treatment of these people represent a cost around of 30 billion dollars per year [1], [2], [3].

In research conducted by Rangel [4] with 181 stroke patients, the sequels cause dependence for trivial activities. The research showed that 49.6% of the patients had moderate to severe dependence and 49.7% had dysphoric or depressive symptoms.

However, if treatment is performed immediately, recovery in the damaged region is possible. For this, some studies indicate that early stroke detection is performed from the analysis of cerebral microbleeds (CMBs), that is a prodromal stage of stroke [5]. Medical images obtained by Computed Tomography (CT) or Methods using Magnetic Resonance Imaging (MRI) are very important to aid diagnose this disease [6]. Although the two exams present detailed images of body parts, each method has different objectives and operating mechanisms, as well dissimilar applications [7]. MRI has higher sensitivity and it has no radiation to patient, different from TC scan exam. However, this method has limitations, due to the longer examination time and the costs, both of which hamper its use on a large scale; consequently, it is usually only found at large hospitals [8]. Otherwise, CT is faster, is less affected by noise and has a lower cost, so is available to a larger number of people. These tests allow the extent and severity of the stroke to be evaluated by evaluating the flow and blood volume [9]. Also, CT is more indicated for lung and bone analysis, in the study of abdominal structures and for initial assessment of the central nervous system [10].

Some characteristics of a CT image can be used to highlight the differences and similarities between the objects within the image. The medical specialist performs a visual segmentation of the exams to try to distinguish particular areas that display pathological changes. However, the changes that appear on CT images are subtle and in some cases imperceptible to a simple visual analysis [11]; moreover, CT scans provide information that are invisible to the human eye.

These digital images usually have approximately 4000 different gray levels. Whereas the Human Visual System (HVS) is able to distinguish only about 64 grayscale for the same condition of lightness. Thus, a computational analysis may reveal a much larger number of attributes that cannot be detected by the human eye.