Patterns and Measurements in Biomedical Engineering
by Associate Professor Dr. Ibrahima Faye
Pattern recognition and objective measurement techniques in biomedical engineering are used to solve practical problems in medicine. The following examples (fruit classification and measurement) explain the concepts of pattern recognition and objective measurement techniques.
Biomedical engineering is a multidisciplinary field with special focus in improving human health. Biomedical engineers construct diagnostic tools and devices used by medical practitioners. For this reason, biomedical engineering incorporates approaches from different research fields (e.g., engineering, mathematics, physics, computer sciences) to come up with practical solutions with minimal constraints from high cost as well as ethical and pain complications.
Pattern recognition and objective measurements are two expressions that can be used to illustrate how biomedical engineering could be applied in medicine. Differentiating between normal and abnormal situations is a fundamental step in medicine. For instance, various pattern recognition techniques are used to detect cancer in breast, lung, skin, etc.
In the same vein, an objective measurement that statistically explains pattern differences helps medical doctors to assess severity of disease and decide appropriate treatments. An effective objective measurement could solve the problem of assessment difference from different doctors (inter-variability), and from a doctor at different times (intra-variability).
FEATURES FOR CLASSIFICATION OF PATTERNS
Consider a bag containing apples and oranges, and one fruit is to be removed from the bag. For each fruit selected from the bag, without seeing it, we need to determine whether it is an orange or an apple based on a given criteria or feature. From the shape and the skin texture, we are able to classify the fruit as apple or orange. Thus, we would say the shape and the skin texture are good features, i.e., features that can be used to differentiate between apples and oranges in the bag. If the weights of oranges and apples are in the same range, however, then we may not be able to classify the fruits correctly. In this case, we would say that the weight is not an optimal feature to distinguish between oranges and apples.
As fruit weight may not be a good identifying feature, we can further use a transformed feature of the fruits as part of the identification process. If each selected fruit is pressed (i.e., ‘transformed’) into juice, the taste of the juice and the texture of the pulp allow us to differentiate between oranges and apples. Pattern recognition is an important technique in detection of breast cancer. The standard diagnostic method is analysis of mammogram images obtained from X-rays. For each patient, two images are obtained for each breast. Radiologists will examine the images to detect any sign of abnormality. The work can be tedious especially in screening programs where a large number of images have to be examined. Computerised system is needed to assist radiologists to identify suspicious mammograms or areas of mammograms where radiologists could focus on.
The main objective of a computer-aided system is to identify abnormality, and finding good features (characteristics) will be the key for this task. Good features are those that are able to differentiate normal from abnormal images using mathematical and statistical features (e.g., mean, standard deviation).
As mentioned previously, the fruits could be differentiated after a transformation. For mammogram images, such transformations are translated into mathematical functions such as Fourier and Wavelet transformations, which allow us to visualise the functions in different dimensions. An image is a mathematical function whereby each point in the image (pixel) corresponds to a grey-level value representing the intensity. Such transformations are useful in feature differentiation.
Consider a scenario where we have to estimate the percentage of yellow parts in red apples. For such a visual-based estimation, different people may give different percentages for the same apple; inter-variability (different percentages given by different people) and intra-variability (different percentages given by the same person) are inevitable owing to the subjectivity of the task.
Thus, an objective optical measurement could be used to scan a fruit and assign a specific colour to each pixel based on a specified rule that is common to all examined fruits. Such an objective measurement of colour percentage not only solves the problem of variability, but also reduces assessment time especially if there is a large number of fruits to be examined.
Computerised system is needed to assist radiologists to identify suspicious mammograms or areas of mammograms where radiologists could focus on
Estimating the healing of an ulcer wound is a challenging task. For chronic ulcers, the healing process can be very long and medical doctors need to assess the effects of prescribed medicines. Because of the pain, a non-invasive method is preferred. An assessment based on doctors’ observation is usually subjected to inter- and intra-variability.
After the wound is treated with a medication for a period of time, the rate of healing needs to be determined based on the volume of wound that has recovered. Volume estimation can be challenging, as it depends on the shape of the organ where the wound is situated. An objective measure obtained using a 3D camera can thus be used, and algorithms can be developed to estimate the recovered volume, and hence, the overall healing.
Similarly, estimation of acne severity is subjective when it is based on visual observation. The assessment for acne includes counting of the number of pimples, their sizes and also the colour. For an objective assessment, a segmentation of the face image is needed. The face is subdivided in regions, which consist of pimple, and non-pimple regions. The first stage of evaluation involves the counting of the number of pimples as well as the estimation of the sizes.
The second stage consists of classification of the pimples depending on their colours. Each pixel in a pimple region can be represented by its content in terms of the three primary colours: Red Blue Green (RGB). A pimple region will then be assigned to a class based on its colour distribution of RGB.
As for conclusion, pattern recognition is used to differentiate between normal and abnormal cases in biomedical applications. For objective measurement technique, there should be very minor differences in the measurements of an object by different operators or the same operator at different times. Finally, finding practical and viable solutions in medicine is a challenging task; an optimal combination of different expertise is the key for developing the right tools.
About the Author
DR. IBRAHIMA FAYE is an Associate Professor at Universiti Teknologi PETRONAS, Tronoh, Malaysia. He is based at the Department of Fundamental and Applied Sciences and the Centre for Intelligent Signal and Imaging Research (CISIR). His BSc, MSc and PhD degrees in Mathematics are from University of Toulouse, France, while his MS degree in Engineering of Medical and Biotechnological data is from Ecole Centrale Paris, France. His research interests include Engineering Mathematics, Signal and Image Processing, Pattern Recognition, and Dynamical Systems. Find out more about Ibrahima at http://www.scientificmalaysian.com/members/ibrahimafaye/