Computer aided detection of small acute intracranial hemorrhage on computer tomography of brain

https://doi.org/10.1016/j.compmedimag.2007.02.010Get rights and content

Abstract

Introduction

Detection of acute intracranial hemorrhage (AIH) is a primary task in image interpretation of computer tomography (CT) of brain for patients suffering from acute neurological disturbance or head injury. Although CT readily depicts AIH, interpretation can be difficult especially when the lesion is inconspicuous or the reader is inexperienced.

Objective

To develop a computer aided detection system that improves diagnostic accuracy of small AIH on brain CT.

Materials and methods

Intracranial contents are first segmented by thresholding and morphological operations, which are then subjected to denoising and adjustment for CT cupping artifacts. The brain is then automatically realigned into normal position. AIH candidates are extracted based on top-hat transformation and left–right asymmetry. AIH candidates are registered against a normalized coordinate system such that the candidates are rendered anatomical information. True AIH is differentiated from mimicking normal variants or artifacts by a knowledge-based classification system incorporating rules that make use of quantified imaging features and anatomical information.

A total of 186 clinical cases, including 62 CT studies showing small (<1 cm) AIH, and 124 controls, were retrospectively collected. Forty positive cases and 80 controls were used for the training of the CAD. Twenty-two positive cases and 44 controls were used in the validation of the CAD system. Regions of AIH identified by two experienced radiologists were used as gold standard. The size of individual AIH volume was also recorded.

Results

On a per patient basis, the system achieved sensitivity of 95% (38/40) and specificity of 88.8% (71/80) in the training dataset. The sensitivity and specificity were 100% (22/22) and 84.1% (37/44) respectively for the diagnosis of AIH in the validation cases.

Individual cases contained variable number of AIH volumes. There were 77 lesions in the 40 training cases and 46 lesions in the 22 validation cases. On a per lesion basis, the sensitivities were 84.4% (65/77) and 82.6% (38/46) for all lesions 10 mm or smaller for the training and validation datasets, respectively. False positive rates were 0.19 (23/120) and 0.29 (19/66) false positive lesion per case for the training and validation datasets, respectively.

Conclusion

This study demonstrated that CAD is valuable for detection of small AIH on brain CT.

Introduction

Acute intracranial hemorrhage (AIH) literally means recent bleeding inside the confine of the skull. It comprises bleeding inside or outside the brain substance, which are termed intraaxial and extraaxial hemorrhage, respectively. Intraaxial hemorrhage can be further specified as to the exact anatomical location of bleeding, e.g. cerebral hemorrhage and brainstem hemorrhage, and intraventricular hemorrhage (IVH). Extraaxial hemorrhage is classified according to the anatomical layer of meninges where bleeding occurs, namely extradural hemorrhage (EDH), subdural hemorrhage (SDH), subarachnoid hemorrhage (SAH).

AIH can be an important cause of acute neurological disturbance, e.g. cerebral hemorrhage causing hemiplegia, or consequence of head injury, e.g. extradural hemorrhage.

Identification of AIH is of crucial clinical significance, because this dictates very different management strategies; however, neither symptoms and signs nor other clinical parameters is accurate in differentiating AIH from other causes of neurological disturbance [1], [2]. CT has been the primary modality for detection of AIH because it is widely available, quick to perform, and readily depicts AIH [3], [4].

Visualization of an acute clot on CT depends on its intrinsic physical properties including the density, volume, location, and relationship to surrounding structures, and technical factors including scanning angle, slice thickness, and windowing [5]. It is conceivable that AIH can become difficult to identify when it is small. Although no formal classification of the size of AIH is known. The long axis diameter and thickness in the transverse plane represent the conventionally measured dimensions of intraaxial hematomas and extraaxial hematomas, respectively. The current study defines a lesion as small if it is (a) intraaxial hemorrhage having a long axis diameter equal or less than 1 cm, or (b) extraaxial hemorrhage having a thickness equal or less than 1 cm. All the sizes quoted in the following discussions refer to either the long axis diameter for intraaxial hematomas or the thickness of extraaxial hematomas measured in the transverse plane.

It is obvious that diagnosis of AIH requires correct interpretation of the demonstrable AIH on CT. This can become difficult when the lesion is inconspicuous, e.g. small or being masked by normal structures, or when the reader is inexperienced.

In most parts of the world, acute care physicians, including emergency physicians, internists, or neural surgeons, are the only ones to read the CT images at odd hours, when radiologists’ expertise may not be immediately available. However, the skill of acute care physicians regarding interpretation of brain CT has been shown to be imperfect [6]. Another study has shown that radiology residents can, albeit infrequently, overlook hemorrhage on brain CT [7].

Even for the best human observers, it has long been recognized that errors in image interpretation, including erroneous perception or analysis, are inevitable [8]. It is envisaged that CAD may help to improve the accuracy in detection of AIH and hence decrease the risk of misdiagnosis and mismanagement. One system has been developed to detect acute middle cerebral artery (ischemic) stroke [9], but there has been no published work in the CAD of AIH to the best of the authors’ knowledge. A few systems have been reported recently by Hodgson et al., Yang et al., and Goto et al. [10], [11], [12]. Yet none of these systems have reported success in detecting small AIH, which are the problematic lesions that could present as challenge to both human observers and computer systems. Success in tackling the small lesions is expected to produce the greatest impact in clinical practice.

Therefore, the objective of this research is to develop a CAD system that identifies small AIH to help in the management of patients suffering from head injury or acute neurological disturbance in an emergency setting.

The current system has been built for the diagnosis of small AIH. It differs from existing CAD products, e.g. malignancy detection in mammography and nodule detection in chest radiograph or CT, in that the system is intended to be used by clinicians other than radiologists and that the system rates the authenticity of candidate lesions in different portions of the image dataset differently, depending on their anatomical positions and imaging features.

As noted before, in emergent settings, expert radiologists may not be readily available to provide the often crucial image interpretation. Therefore, the duty is shifted to clinicians who may not be best equipped for the task. It is therefore believed that CAD may become useful in these situations, in addition to its proven value for screening examinations. Special considerations need to be made because observers of less expertise may not be confident or knowledgeable enough to judge the correctness of CAD outputs. Therefore, CAD systems targeted for non-radiologists need to minimize the false positive rates.

During the development stage of the system, it was found that the myriad combinations of imaging features of AIH in different parts of the brain could not be adequately described without reference to the anatomical positions where the lesions are found. But when the candidate lesions are divided up based on the anatomical position, classification between genuine AIH and mimicking variants or artifacts become feasible. This contrasts against target lesions of many CAD systems which are well-described with relatively little variation in their configurations, which are hence less dependent on the anatomical information in comparison to the local imaging features.

Section snippets

Materials

One hundred and eighty-six brain CT studies, including 62 cases showing AIH and 124 cases showing no AIH, were retrospectively retrieved from the CT archive of the Princess Margaret Hospital in Hong Kong. All were cases performed on an emergency setting for evaluation of head injury or acute neurological disturbance. The studies were anonymized apart from the sex and age. Institute review board approval has been obtained for this study.

All studies were acquired with a single detector CT scanner

Performance in anatomical localization

This method correctly located the level of the floor of middle cranial fossa in 97.5% (117/120) and 95.3% (61/64) of the training and validation cases, respectively. All the other cases were off by one axial section only.

The mid-sagittal planes (MSP) were accurately localized in 69.1% (83/120) and 65.6% (42/64) of the training and validation dataset, respectively, which are defined as system output that is within 1 mm of displacement and 1° of rotation from the genuine MSP. In 22.5% (27/120) and

Detection of small AIH

There is no objective method of classifying AIH according to its size. The target size of detecting AIH smaller than 10 mm is arbitrarily chosen. The width rather than the length is chosen for extraaxial hemorrhage because it is the dimension which is clinically relevant and customarily reported. Although it does imply that 10 mm extraaxial AIH would be larger in area/volume than a 10 mm intraaxial AIH described in this study, it is believed that this convention better reflect the radiologists’

Conclusion

A CAD system capable of identifying small intracranial hemorrhage has been developed. It classifies genuine AIH from mimicking normal variants or artifacts based on both image features and anatomical information, which is made possible by construction of a coordinate system that incorporates positional information of normal brain structures. It is expected that this system can benefit patient care especially in emergency situations when timely management decision need to be made by acute care

Dr. Tao Chan is a radiologist working in Hong Kong. He received his medical education from the Chinese University of Hong Kong. He subsequently completed radiology training in Hong Kong and has become a fellow of the Royal College of Radiologists (UK). Dr. Chan has also received fellowship training in interventional radiology in both Hong Kong and Australia. He is currently a PhD candidate in Department of Health Technology and Informatics at the Hong Kong Polytechnic University. His has been

References (22)

  • R. Hodgson

    CAD system for detecting haemorrhage in CT of stroke

  • Cited by (128)

    • Entropy based automatic unsupervised brain intracranial hemorrhage segmentation using CT images

      2022, Journal of King Saud University - Computer and Information Sciences
    • An efficient stacked bidirectional GRU-LSTM network for intracranial hemorrhage detection

      2024, International Journal of Imaging Systems and Technology
    View all citing articles on Scopus

    Dr. Tao Chan is a radiologist working in Hong Kong. He received his medical education from the Chinese University of Hong Kong. He subsequently completed radiology training in Hong Kong and has become a fellow of the Royal College of Radiologists (UK). Dr. Chan has also received fellowship training in interventional radiology in both Hong Kong and Australia. He is currently a PhD candidate in Department of Health Technology and Informatics at the Hong Kong Polytechnic University. His has been working on the concepts and software implementation of Computer-aided Diagnosis in acute intracranial hemorrhage since 2004.

    View full text