Image Processing Using Feature-Based Segmentation Techniques for the Analysis of Medical Images ^†

Abstract

Image segmentation is a fundamental task in computer vision in which an image is divided into many regions or segments, each of which corresponds to a separate object or part of an item within the image. Image segmentation’s major purpose is to simplify an image’s representation for analysis and interpretation, making it easier for a computer to comprehend and extract meaningful information from visual data. Adaptive K-means clustering is a variant of the classic K-means clustering algorithm in which the number of clusters (K) is continuously adjusted during the clustering process. Unlike classic K-means, which requires you to choose the number of clusters before executing the algorithm, adaptive K-means identifies the best number of clusters based on the features of the data. The proposed model works as follows. Firstly, pre-processing is performed by acquiring all the input images. Secondly, adaptive k-means clustering is employed for segmentation. Thirdly, important features are automatically extracted from X-ray images by making use of a feature-based image registration technique. Then, the detection of bone fractures is automatically carried out. The results are compared with those of existing studies, and it is observed that this model provides better results.

1. Introduction

A significant amount of research is being carried out using various image processing and machine learning methods to analyze and predict complex health problems in the medical domain. In medical IoT systems, there is a significant increase in technology day-by-day, and health varies in the daily life cycle of a human being. The use of the IoT is being explored in various applications; this includes the various roles of image processing [1,2]. Medical IoT smart systems include important clinical aspects like remote patient monitoring, providing medical services online, and predicting various critical health issues based on preventive parameters as part of patient IoT health care systems. By varying preventive parameters, IoT sensing smart computational devices and smart medical equipment can be used to monitor the health of patients by detecting bone diseases, Spondylosis, and bone fractures in early stages. At present, both young and elderly people are facing many bone-relevant diseases or infections, so IoT-based sensing smart computational devices and smart medical equipment are required for early diagnosis to provide accurate and efficient results [3].

Medical photographs are an important part of modern healthcare since they aid in the diagnosis, treatment, and monitoring of numerous medical diseases. Each image type contains distinct information and is utilised for specialized diagnostic purposes [4,5,6,7]. Medical photographs are a valuable diagnostic tool for healthcare providers. They enable clinicians to see parts within the human body, such as bones, organs, tissues, and blood vessels, in order to diagnose abnormalities, injuries, diseases, and other medical disorders [8,9,10].

In recent years, medical imaging has shifted from traditional film-based technologies to digital imaging, which provides various benefits such as easier storage, faster image capture, and the ability to electronically exchange images for remote consultation. The acquisition of three-dimensional (3D) and four-dimensional (4D) medical images has been made possible by advanced technologies, allowing for more detailed and dynamic visualizations of anatomical structures and physiological processes [10,11,12].

PACS stands for Picture Archiving and Communication System. PACS is a computerized system for storing, retrieving, and managing medical pictures. It gives healthcare practitioners easy access to a patient’s entire imaging history [13,14]. Examples of X-ray images are shown in Figure 1.

2. Related Works

In the past, examining and identifying diseases in medical field was a lengthy process. Image segmentation techniques are used in medical images to study the internal structures of the human body, detect tumors, identify bone fractures, and estimate tumor size, etc. Segmentation for X-ray images is a challenging task because they are complex, the quality is low, and they are affected by noise. So, extracting the region of interest and boundaries from X-ray images is a challenging task. Also, they vary in resolution, orientation, and intensity based on the equipment used to capture the image. These factors will greatly affect the result of the segmentation process. Another problem with bone X-ray images is that bone regions overlap with other organs and muscles. The joints between bones must also be considered in the segmentation process for better analysis [15,16].

The process of breaking an image into many separate parts or segments, each of which corresponds to a relevant object, region, or section of an image, is known as image segmentation [17]. Image segmentation’s major purpose is to simplify an image’s representation for analysis and interpretation, making it easier for a computer to comprehend and extract important information from visual data.

Image segmentation techniques and algorithms include, but are not limited to, thresholding, region-based methods, edge-based methods, clustering methods (such as K-means), and deep learning-based approaches (such as convolutional neural networks). Image segmentation can be difficult, especially when dealing with complicated or congested photos, fluctuating lighting conditions, or irregularly shaped objects [18]. Furthermore, choosing the correct segmentation method and parameters is critical for attaining accurate results. Regarding metrics for evaluation, image segmentation quality can be measured using a variety of measures, including the Dice coefficient, the Jaccard index, and boundary-based metrics. These metrics aid in quantifying the resemblance of segmented regions to ground truth data [19].

To increase the quality of the results, preprocessing processes like noise reduction, picture enhancement, and color space conversion are frequently used before segmentation. To refine the segmentation result, post-processing processes such as region merging or splitting, boundary smoothing, and filling small gaps can be used. Regarding interactive segmentation, when dealing with difficult images, some segmentation jobs may benefit from human participation in which the user offers input to direct the segmentation process.

3. Methodology

The workflow of the recommended model is shown in Figure 2.

A one-dimensional Gaussian distribution is represented below in Equation (1):

$$f\left(x\right)=\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$\sqrt{2\ast \mathsf{\pi }\ast \mathsf{\sigma }}$}\right.e^{\frac{-x^2}{2{\ast \mathsf{\sigma }}^2}}$$

(1)

A two-dimensional Gaussian distribution is represented below in Equation (2):

$$f\left(x,y\right)=\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$\sqrt{2\ast \mathsf{\pi }\ast {\mathsf{\sigma }}^2}$}\right.e^{\frac{-x^2+y^2}{2\ast {\mathsf{\sigma }}^2}}$$

(2)

This method automatically generates accurate segmentation results by performing a simple operation considering the input k-value. Without mentioning the seed value or the number of clusters in the iteration manner, the k-means algorithm automatically generates the seed value or number of clusters in less time. The step-by-step procedure for the adaptive k-means clustering method is given below:

Step 1: Input a k-value as clusters where k = 1, 2, 3, … N.

Step 2: Assign k-data points randomly to any of the k-cluster.

Step 3: Find the distance between the initial center points of clusters by applying Euclidian distance, as shown below in Equation (3):

$$\mathrm{ED}=\sqrt{({\left(\mathrm{a}1-\mathrm{b}1\right)}^2+{(\mathrm{a}2-\mathrm{b}2)}^2}$$

(3)

where a1, a2, b,1 and b2 are points on a plane.

Step 4: Based on data points from various centers to each cluster, we calculate the distance of the data points, and depending on the distance value achieved, we reassign all data points to nearest cluster.

Step 5: A new center cluster value is calculated, and the steps are repeated until the number of iterations is assigned.

The error rate is calculated as given below in Equation (4):

$$\mathrm{A}=\frac{\left|u_b-u_c\right|}{u_c}$$

(4)

where A is the percentage error, $u_b$ is the observed actual value, and $u_c$ is the expected value.

Entropy is calculated as given below in Equation (5):

$$\mathrm{E}=\mathrm{s}\mathrm{u}\mathrm{m}\left(\mathrm{q}\ast {\mathrm{l}\mathrm{o}\mathrm{g}}_2\left(\mathrm{q}\right)\right)$$

(5)

where q is the count of the normalized histogram.

Accuracy is calculated as given below in Equation (6):

$$A=\frac{\mathrm{T}\mathrm{P}\mathrm{s}+\mathrm{T}\mathrm{N}\mathrm{s}}{\mathrm{T}\mathrm{P}\mathrm{s}+\mathrm{T}\mathrm{N}\mathrm{s}+\mathrm{F}\mathrm{P}\mathrm{s}+\mathrm{F}\mathrm{N}\mathrm{s}}$$

(6)

where TPs is a true positive, TNs is a true negative, FPs is a false positive, and FNs is a false negative.

4. Results and Discussion

The performance of the results is demonstrated in this Results and Discussion section. The input X-ray images were taken from the website kaggle, and the results were obtained using MATLAB R2019a. The results are shown in Figure 3 and Figure 4:

The performance measures are shown below in

Table 1. Performance measures.

Performance Measures	Results 1	Results 2
Input image	X-ray—hand	X-ray—leg and ankle
Error rate	0.00056	0.00048
Entropy	1.0	0.98
Elapsed time	1.21 s	1.47 s
Accuracy	92%	91%

.

As per our proposed system, we achieved good accuracy results of 92% by using our proposed system.

5. Conclusions

Image segmentation is a key component of computer vision that has numerous practical applications in industries such as healthcare, autonomous driving, and remote sensing. The method of segmentation used is determined based on the work at hand, the data, and the amount of detail sought. Adaptive K-means clustering is a version of the conventional K-means clustering technique that continuously adjusts the number of clusters (K) during the clustering process. Unlike traditional K-means, which requires you to choose the number of clusters before using the algorithm, adaptive K-means determines the appropriate number of clusters depending on the properties of the data. The proposed model works as follows. First, the input photos are acquired and pre-processed. Second, for segmentation, adaptive K-means clustering is used. Third, key characteristics are retrieved automatically from X-ray images using a feature-based image registration technique. Finally, the early detection of bone fractures is then performed automatically.