Towards edge devices implementation: deep learning model with visualization for COVID-19 prediction from chest X-ray.

Due to the outbreak of COVID-19 disease globally, countries around the world are facing shortages of resources (i.e. testing kits, medicine). A quick diagnosis of COVID-19 and isolating patients are crucial in curbing the pandemic, especially in rural areas. This is because the disease is highly contagious and can spread easily. To assist doctors, several studies have proposed an initial detection of COVID-19 cases using radiological images. In this paper, we propose an alternative method for analyzing chest X-ray images to provide an efficient and accurate diagnosis of COVID-19 which can run on edge devices. The approach acts as an enabler for the deep learning model to be deployed in practical application. Here, the convolutional neural network models which are fine-tuned to predict COVID-19 and pneumonia infection from chest X-ray images are developed by adopting transfer learning techniques. The developed model yielded an accuracy of 98.13%, sensitivity of 97.7%, and specificity of 99.1%. To highlight the important regions in the X-ray images which directs the model to its decision/prediction, we adopted the Gradient Class Activation Map (Grad-CAM). The generated heat maps from the Grad-CAM were then compared with the annotated X-ray images by board-certified radiologists. Results showed that the findings strongly correlate with clinical evidence. For practical deployment, we implemented the trained model in edge devices (NCS2) and this has achieved an improvement of 90% in inference speed compared to CPU. This shows that the developed model has the potential to be implemented on the edge, for example in primary care clinics and rural areas which are not well-equipped or do not have access to stable internet connections.

Dynamic sampling of images from various categories for classification based incremental deep learning in fog computing.

Incremental learning evolves deep neural network knowledge over time by learning continuously from new data instead of training a model just once with all data present before the training starts. However, in incremental learning, new samples are always streaming in whereby the model to be trained needs to continuously adapt to new samples. Images are considered to be high dimensional data and thus training deep neural networks on such data is very time-consuming. Fog computing is a paradigm that uses fog devices to carry out computation near data sources to reduce the computational load on the server. Fog computing allows democracy in deep learning by enabling intelligence at the fog devices, however, one of the main challenges is the high communication costs between fog devices and the centralized servers especially in incremental learning where data samples are continuously arriving and need to be transmitted to the server for training. While working with Convolutional Neural Networks (CNN), we demonstrate a novel data sampling algorithm that discards certain training images per class before training even starts which reduces the transmission cost from the fog device to the server and the model training time while maintaining model learning performance both for static and incremental learning. Results show that our proposed method can effectively perform data sampling regardless of the model architecture, dataset, and learning settings.

Objective assessment of vitiligo with a computerised digital imaging analysis system.

An objective tool to quantify treatment response in vitiligo is currently lacking. This study aimed to objectively evaluate the treatment response in vitiligo by using a computerised digital imaging analysis system (C-DIAS) and to compare it with the physician's global assessment (PGA). Tacrolimus ointment 0.1% (Protopic; Astellas Pharma Tech,Toyama, Japan) was applied twice daily on selected lesions which were photographed every 6 weeks for 24 weeks. The primary efficacy end-point was the mean percentage of repigmentation (MPR), as assessed by the digital method (MPR-C-DIAS) or by the PGA. The response was categorised into none (0%), mild (1-25%), moderate (26-50%), good (51-75%) and excellent (76-100%). MPR-C-DIAS: Out of 56 patients, 44 (79%) responded. Overall, the response was mild in 22 (39%), moderate in 21(40%) and good in one (2%) patient(s). A total of 39 (70%) patients responded as measured by PGA. The repigmentation was mild in 27(48%), moderate in 10 (18%) and good to excellent in two (4%) patients. The κ test of consistency was 0.17 (P = 0.053), which shows poor agreement between the two assessment methods, although this is not statistically significant. The C-DIAS can be used to perform an objective analysis of repigmentation or depigmentation in vitiligo skin lesions in response to treatment.

3D surface roughness measurement for scaliness scoring of psoriasis lesions.

Psoriasis is an incurable skin disorder affecting 2-3% of the world population. The scaliness of psoriasis is a key assessment parameter of the Psoriasis Area and Severity Index (PASI). Dermatologists typically use visual and tactile senses in PASI scaliness assessment. However, the assessment can be subjective resulting in inter- and intra-rater variability in the scores. This paper proposes an assessment method that incorporates 3D surface roughness with standard clustering techniques to objectively determine the PASI scaliness score for psoriasis lesions. A surface roughness algorithm using structured light projection has been applied to 1999 3D psoriasis lesion surfaces. The algorithm has been validated with an accuracy of 94.12%. Clustering algorithms were used to classify the surface roughness measured using the proposed assessment method for PASI scaliness scoring. The reliability of the developed PASI scaliness algorithm was high with kappa coefficients>0.84 (almost perfect agreement).

Computerised image analysis of vitiligo lesion: evaluation using manually defined lesion areas.

BACKGROUND: Vitiligo is a cutaneous pigmentary disorder characterized by depigmented macules and patches that result from loss of epidermal melanocytes. Physician evaluates the efficacy of treatment by comparing the extent of vitiligo lesions before and after treatment based on the overall visual impression of the treatment response. This method is called the physician's global assessment (PGA) which is subjective. In this article, we present an innovative digital image processing method to determine vitiligo lesion area in an objective manner. METHOD: The digital method uses Independent Component Analysis (ICA) to generate melanin-based images representing skin areas due to melanin followed by Region Growing process to segment vitiligo lesion from normal skin. RESULTS: Based on 41 digital images of vitiligo lesions taken from 18 patients, the proposed method achieved sensitivities of 0.9105 ± 0.0161, specificities of 0.9973 ± 0.0009 and accuracies of 0.9901 ± 0.0028 at 95% confidence level. CONCLUSION: With the proposed method, physicians are able to assess vitiligo treatment efficacies objectively.

Body surface area measurement and soft clustering for PASI area assessment.

Psoriasis is a common skin disorder with a prevalence of 0.6 - 4.8% around the world. The most common is plaques psoriasis and it appears as red scaling plaques. Psoriasis is incurable but treatable in a long term treatment. Although PASI (Psoriasis Area and Severity Index) scoring is recognised as gold standard for psoriasis assessment, this method is still influenced by inter and intra-rater variation. An imaging and analysis system called α-PASI is developed to perform PASI scoring objectively. Percentage of lesion area to the body surface area is one of PASI parameter. In this paper, enhanced imaging methods are developed to improve the determination of body surface area (BSA) and lesion area. BSA determination method has been validated on medical mannequin. BSA accuracies obtained at four body regions are 97.80% (lower limb), 92.41% (trunk), 87.72% (upper limb), and 83.82% (head). By applying fuzzy c-means clustering algorithm, the membership functions of lesions area for PASI area scoring have been determined. Performance of scoring result has been tested with double assessment by α-PASI area algorithm on body region images from 46 patients. Kappa coefficients for α-PASI system are greater than or equal to 0.72 for all body regions (Head - 0.76, Upper limb - 0.81, Trunk - 0.85, Lower limb - 0.72). The overall kappa coefficient for the α-PASI area is 0.80 that can be categorised as substantial agreement. This shows that the α-PASI area system has a high reliability and can be used in psoriasis area assessment.

Analysis of retinal fundus images for grading of diabetic retinopathy severity.

Diabetic retinopathy (DR) is a sight threatening complication due to diabetes mellitus that affects the retina. In this article, a computerised DR grading system, which digitally analyses retinal fundus image, is used to measure foveal avascular zone. A v-fold cross-validation method is applied to the FINDeRS database to evaluate the performance of the DR system. It is shown that the system achieved sensitivity of >84%, specificity of >97% and accuracy of >95% for all DR stages. At high values of sensitivity (>95%), specificity (>97%) and accuracy (>98%) obtained for No DR and severe NPDR/PDR stages, the computerised DR grading system is suitable for early detection of DR and for effective treatment of severe cases.

Analysis of foveal avascular zone in colour fundus images for grading of diabetic retinopathy severity.

Diabetic retinopathy (DR) is a sight threatening complication due to diabetes mellitus that affects the retina. At present, the classification of DR is based on the International Clinical Diabetic Retinopathy Disease Severity. In this paper, FAZ enlargement with DR progression is investigated to enable a new and an effective grading protocol DR severity in an observational clinical study. The performance of a computerised DR monitoring and grading system that digitally analyses colour fundus image to measure the enlargement of FAZ and grade DR is evaluated. The range of FAZ area is optimised to accurately determine DR severity stage and progression stages using a Gaussian Bayes classifier. The system achieves high accuracies of above 96%, sensitivities higher than 88% and specificities higher than 96%, in grading of DR severity. In particular, high sensitivity (100%), specificity (>98%) and accuracy (99%) values are obtained for No DR (normal) and Severe NPDR/PDR stages. The system performance indicates that the DR system is suitable for early detection of DR and for effective treatment of severe cases.

Determination of skin repigmentation progression.

In this paper, we describe an image processing scheme to analyze and determine areas of skin that have undergone repigmentation in particular, during the treatment of vitiligo. In vitiligo cases, areas of skin become pale or white due to the lack of skin pigment called melanin. Vitiligo treatment causes skin repigmentation resulting in a normal skin color. However, it is difficult to determine and quantify the amount of repigmentation visually during treatment because the repigmentation progress is slow and moreover changes in skin color can only be discerned over a longer time frame typically 6 months. Here, we develop a digital image analysis scheme that can identify and determine vitiligo skin areas and repigmentation progression on a shorter time period. The technique is based on principal component analysis and independent component analysis which converts the RGB skin image into a skin image that represent skin areas due to melanin and haemoglobin only, followed by segmentation process. Vitiligo skin lesions are identified as skin areas that lack melanin (non-melanin areas). In the initial studies of 4 patients, the method has been able to quantify repigmentation in vitiligo lesion. Hence it is now possible to determine repigmentation progression objectively and treatment efficacy on a shorter time cycle.