The bodily movements of a person can be heavily restricted because of a spinal cord disorder. These disorders could be a result of various internal or external factors but it is believed that a preliminary diagnostic for the magnitude of the spinal dysfunction can be determined by the movement of the patient itself. The behavioural patterns of a person doing specific tasks with spinal dysfunction varies from someone who does the same tasks with a healthy back. In this contribution, it is demonstrated that this difference can be generalised and then implemented to predict the spinal health index of the patients with only but an input video of them performing predetermined tasks. A novel technique of abstracting dynamic heat-map as the sequential input to the custom LSTM model is featured allowing to outperform previous techniques by reaching 98.4% validation accuracy with a realtime computational complexity of ≤11 milliseconds.

Dynamic Pose Diagnosis with BlazePose and LSTM for Spinal Dysfunction Risk Estimation

The bodily movements of a person can be heavily restricted because of a spinal cord disorder. These disorders could be a result of various internal or external factors but it is believed that a preliminary diagnostic for the magnitude of the spinal dysfunction can be determined by the movement of the patient itself. The behavioural patterns of a person doing specific tasks with spinal dysfunction varies from someone who does the same tasks with a healthy back. In this contribution, it is demonstrated that this difference can be generalised and then implemented to predict the spinal health index of the patients with only but an input video of them performing predetermined tasks. A novel technique of abstracting dynamic heat-map as the sequential input to the custom LSTM model is featured allowing to outperform previous techniques by reaching 98.4% validation accuracy with a realtime computational complexity of ≤11 milliseconds. 

In clinical routine, a number of radiation techniques are used to diagnose physical health conditions. The most common are CT-Scan and X-Ray techniques. Each of the techniques has its pros and cons. For example, CT-Scan is often used to gain a very clear composition of the organs under inspection. However, a person may suffer from a large volume of radiation exposure. Dual Energy X-ray Absorptiometry - DXA is often an alternative to CT-Scan. This is due to the fact that the DXA technique diffuses low doses of radiation. As a result, an image generated with a DXA technique often has less information than those generated by a CT-Scan technique. However, DXA's role in sarcopenia diagnosis is emerging. In this respect, the detection of muscle mass changes in sarcopenia diagnosis requires follow up treatment. Thus, there is a need for a method to improve the DXA produced images that can be used to help identify the muscle area more accurately. This paper proposes an approach to enhance the original DXA image to make it more appropriate to support an expert's diagnosis. The accuracy results measured with JI, CM, DS, and HD demonstrates that the proposed method outperformed other methods with the results on each class as follows: Class A: 0.68, 0.75, 0.81, and 17.61 respectively; Class B: 0.62, 0.64, 0.76, and 16.99 respectively; Class C: 0.52, 0.48, 0.68, and 18.75 respectively. 

Hybrid Area-based Techniques for Muscle Mass Identification on Dual Energy X-Ray Absorptiometry Images

In clinical routine, a number of radiation techniques are used to diagnose physical health conditions. The most common are CT-Scan and X-Ray techniques. Each of the techniques has its pros and cons. For example, CT-Scan is often used to gain a very clear composition of the organs under inspection. However, a person may suffer from a large volume of radiation exposure. Dual Energy X-ray Absorptiometry - DXA is often an alternative to CT-Scan. This is due to the fact that the DXA technique diffuses low doses of radiation. As a result, an image generated with a DXA technique often has less information than those generated by a CT-Scan technique. However, DXA's role in sarcopenia diagnosis is emerging. In this respect, the detection of muscle mass changes in sarcopenia diagnosis requires follow up treatment. Thus, there is a need for a method to improve the DXA produced images that can be used to help identify the muscle area more accurately. This paper proposes an approach to enhance the original DXA image to make it more appropriate to support an expert's diagnosis. The accuracy results measured with JI, CM, DS, and HD demonstrates that the proposed method outperformed other methods with the results on each class as follows: Class A: 0.68, 0.75, 0.81, and 17.61 respectively; Class B: 0.62, 0.64, 0.76, and 16.99 respectively; Class C: 0.52, 0.48, 0.68, and 18.75 respectively.

X-ray images of the lateral spine are important for diagnosing spine problems such as osteoporosis, bone fractures, and spondylosis. In order to identify bone diseases, often a series of images is required. These are taken using a low level of X-ray radiation to reduce the risk of exposure to overshoot radiation. Dual Energy X-ray Absorptiometry is a standard medical tool used to diagnose bone diseases. In addition, the spine alignment of each individual person is different others. Therefore, developing an approach that can identify the spine area is challenging. In this work, the algorithm for automatic identification of spine and vertebral bodies is proposed. The proposed method consists of three main steps. The first step, Bi-Histogram Equalization with adaptive sigmoid functions (BEASF), is a technique used for enhancing the spinal and vertebral bodies. In the second step, Density-based and Ellipse-like techniques are combined to locate the curve of the spine. For the third step, object improvement techniques are applied to predict the location of vertebral bodies. The experimental results show that the approach reached 79.67% of Area Overlap Ratio. 81.67% of the Precision value.

Region of Interest Identification on Low-Resolution Lateral Spine Radiography Image using Density-based and Ellipse-like Method

In this research, a new shinbone interosseous space segmentation method was proposed. The X-ray images were obtained from Dual Energy X-ray Absorptiometry (DXA) and consisted of 3 components (muscle, fat, and bone). The DXA scanner produces two different images, gray image and a color image each containing different information. Muscle and fat are two components that can be used to calculate muscle mass. The bone area is used to compute an estimate of the bone mineral density (BMD) measurement as an osteoporosis indicator. Muscle mass is used for the body mass index calculation. X-ray images are the main source of information to measure muscle, fat, and bone area in the human body. The main problem is the ambiguous outline of each component, the position of the legs being placed into the X-ray machine, and the variety of leg shapes. Traditional segmentation methods such as watershed transformation, Noise cancellation, and Image Enhancement were used in preprocessing the images. Analysis of light concentration fluctuations to determine the location of the Gradient Vector Flow (GVF) was the technique used to find the region of interest (ROI), shinbone interosseous space. The Jaccard Index percentage reach 93.92%. Overlapped Area percentage reaches 93.51%.

Segmentation of Shinbone Interosseous Space using GVF Techniques

Dual Energy X-ray Absorptiometry (DEXA) images can be obtained by using low radiation, so it’s safer for patients. An automatic image vertebra pose segmentation can help to identify the disorder of the spine. But DEXA images are low-quality and noisy images, so it’s hard to work with. This paper proposed a method to label vertebrae edges. The proposed method consists of 3 parts. First, preprocessing by using an anisotropic diffusion to reduced noise but preserved an edge. Second, segmentation by using a gradient to identify an edge. Finally, cleansing by using morphological operation and principal component analysis to clean unwanted information. The output of this algorithm is a spine image that labeled edges of the lumbar with 84.14% accuracy, 87.01% recall, 96.22% precision, and 12.55% false negative.

John Gatewood Ham

Papers

Region of Interest Identification on Low-Resolution Lateral Spine Radiography Image using Density-based and Ellipse-like Method

Segmentation of Shinbone Interosseous Space using GVF Techniques

Vertebrae Pose Segmentation based on Temporal Anisotropic Diffusion and Ensembled Gradient