Efficient pre-processing of USF and MIAS mammogram images

Journal of Computer Science, Feb, 2007 by Ayman A. AbuBaker, R.S. Qahwaji, Musbah J. Aqel, Hussam Al-Osta, Mohmmad H. Saleh

Abstract: High quality mammogram images are high resolution and large size images. Processing these images require high computational capabilities. The transmission of these images over the net is sometimes critical especially if the diagnosis of remote radiologists is required. In this paper, a preprocessing technique for reducing the size and enhancing the quality of USF and MIAS mammogram images is introduced. The algorithm analyses the mammogram image to determine if 16-bit to 8-bit conversion process is required. Enhancement is applied later followed by a scaling process to reduce the mammogram size. The performances of the algorithms are evaluated objectively and subjectively. On average 87% reduction in size is obtained with no loss of data at the breast region.

Keywords: Breast cancer, image processing, image reduction, mammogram image

INTRODUCTION

Early detection is the best way to improve breast cancer prognosis since the causes of the disease are still unknown. Breast cancer is the second most prevalent cancer among women after skin cancer [1]. In addition, it accounts for most cancer deaths coming only second to lung cancer [1]. Currently, three methods are used for breast cancer diagnosis: mammography, fine needle aspirate and surgical biopsy. Mammography has a reported malignant sensitivity which varies between 68 and 79% [2]. Fine needle aspirate depends on extracting fluids from a breast lump and inspecting it under the microscope. This method has a reported sensitivity varying from 65 to 98% [2]. Surgical biopsy is more evasive and costly but it is the only test that can confirm malignancy. Efficient machine learning algorithms can enhance the performance of mammogram analysis and provide an equivalent performance in terms of robustness and accuracy for surgical biopsy without its evasiveness and cost.

Mammographic screening allows early detection of non-palpable, non-invasive and early invasive tumors. Hence, it can reduce the mortality from breast cancer by 20-30% [3]. There is an increasing need for automatic and accurate detection of cancer cells. However, the low contrast between the breast cancer cells and normal cells increases the difficulty of early detection.

Most of the work in mammography aims at detecting one or more of the three abnormal structures in mammograms [4]: microcalcifications [5], circumscribed masses [6] and speculated lesions [7]. Other methods depend on classifying the breast lesions as benign or malignant [8]. There are problems with the subjective analysis of mammographic images by radiologist. Subjective analysis depends mainly of the experience of the human operator, but it is also affected by fatigue and other human-related factors. In addition, the interpretation is a repetitive task that requires lot of attention to minute details. Hence, it requires lot of staff time and effort, which results in slowing the diagnosis time. On the other hand, the objective analysis of mammograms, which is carried out by automated systems, provides consistent performance but its accuracy is usually lower. Due to the sensitivity of this problem, we believe that radiologists should be involved and computers should not replace them completely. However, computer systems can help them perform better by enhancing the quality of images, highlighting the suspicious regions and providing better analysis tools.

Most mammogram images are large size and high resolution images that require specialized computing facilities to enables efficient processing. To facilitate the transmission of these images over computer networks image compression techniques are usually applied. In this paper, we present a size reduction algorithm that can be implemented on most mammogram images as a pre-processing step to reduce their size without affecting their quality.

Digitized mammography techniques: There have been various advancements in digital image processing in the fields of filtering, enhancement, segmentation and others. However, the usefulness of the new techniques depends mainly on two important parameters: the spatial and grey-level resolutions [9]. An efficient algorithm should provide a diagnostic accuracy in digital images equivalent to that of conventional films. Pixel size and pixel depth are important factors that could critically affect the visibility of small-low contrast objects, which may carry significant information for diagnosis [10]. Therefore, digital image recording systems for medical purposes must provide high spatial resolution and high contrast sensitivity. Nevertheless, this requirement retards the implementation of digital technologies due to the increment in processing and transmission time, storage capacity and cost. For instance, it has been shown that isolated clusters of microcalcifications are one of the most frequent radiological features of asymptomatic breast cancer [10]. A careful search for the clustered microcalcifications that may herald an early-stage cancer should be carried out on all mammograms [11]. Microcalcifications frequently appear as small-size low-contrast radiopacities [12]. Due to this, a typical mammogram must be digitized at a resolution of approximately 4000 x 5000 pixels with 50- [square]m spot size and 12 or 16 bits, resulting in approximately 30 to 40Mb of digital data. Processing or transmission time of such digital images could be quite long. Archiving the amount of data generated in any screening mammography program also becomes an expensive and difficult challenge [13]. It is clear that advances in technologies for transmission or storage are not sufficient to solve this problem. An efficient data-compression or reduction scheme to reduce the digital data without significant degradation of the medical image quality for human and machine interpretation is needed. Several lossless and lossy compression methods have been investigated for medical imaging applications [14,15].