Deep learning algorithms have become the first choice as an approach to medical image analysis, face recognition, and emotion recognition. In this survey, several deep-learning-based approaches applied to breast cancer, cervical cancer, brain tumor, colon and lung cancers are studied and reviewed. Deep learning has been applied in almost all of the imaging modalities used for cervical and breast cancers and MRIs for the brain tumor. The result of the review process indicated that deep learning methods have achieved state-of-the-art in tumor detection, segmentation, feature extraction and classification. As presented in this paper, the deep learning approaches were used in three different modes that include training from scratch, transfer learning through freezing some layers of the deep learning network and modifying the architecture to reduce the number of parameters existing in the network. Moreover, the application of deep learning to imaging devices for the detection of various cancer cases has been studied by researchers affiliated to academic and medical institutes in economically developed countries; while, the study has not had much attention in Africa despite the dramatic soar of cancer risks in the continent.

https://www.mdpi.com/2313-433X/6/11/121/pdf

Deep Learning in Selected Cancers' Image Analysis-A Survey.

Background Image segmentation is an essential step in the analysis and subsequent characterisation of brain tumours through magnetic resonance imaging. In the literature, segmentation methods are empowered by open-access magnetic resonance imaging datasets, such as the brain tumour segmentation dataset. Moreover, with the increased use of artificial intelligence methods in medical imaging, access to larger data repositories has become vital in method development. Purpose To determine what automated brain tumour segmentation techniques can medical imaging specialists and clinicians use to identify tumour components, compared to manual segmentation. Methods We conducted a systematic review of 572 brain tumour segmentation studies during 2015–2020. We reviewed segmentation techniques using T1-weighted, T2-weighted, gadolinium-enhanced T1-weighted, fluid-attenuated inversion recovery, diffusion-weighted and perfusion-weighted magnetic resonance imaging sequences. Moreover, we assessed physics or mathematics-based methods, deep learning methods, and software-based or semi-automatic methods, as applied to magnetic resonance imaging techniques. Particularly, we synthesised each method as per the utilised magnetic resonance imaging sequences, study population, technical approach (such as deep learning) and performance score measures (such as Dice score). Statistical tests We compared median Dice score in segmenting the whole tumour, tumour core and enhanced tumour. Results We found that T1-weighted, gadolinium-enhanced T1-weighted, T2-weighted and fluid-attenuated inversion recovery magnetic resonance imaging are used the most in various segmentation algorithms. However, there is limited use of perfusion-weighted and diffusion-weighted magnetic resonance imaging. Moreover, we found that the U-Net deep learning technology is cited the most, and has high accuracy (Dice score 0.9) for magnetic resonance imaging-based brain tumour segmentation. Conclusion U-Net is a promising deep learning technology for magnetic resonance imaging-based brain tumour segmentation. The community should be encouraged to contribute open-access datasets so training, testing and validation of deep learning algorithms can be improved, particularly for diffusion- and perfusion-weighted magnetic resonance imaging, where there are limited datasets available.

https://journals.sagepub.com/doi/pdf/10.1177/20552076221074122

Magnetic resonance image-based brain tumour segmentation methods: A systematic review

Glioma is the most aggressive and dangerous primary brain tumor with a survival time of less than 14 months. Segmentation of tumors is a necessary task in the image processing of the gliomas and is important for its timely diagnosis and starting a treatment. Using 3D U-net architecture to perform semantic segmentation on brain tumor dataset is at the core of deep learning. In this paper, we present a unique cloud-based 3D U-Net method to perform brain tumor segmentation using BRATS dataset. The system was effectively trained by using Adam optimization solver by utilizing multiple hyper parameters. We got an average dice score of 95% which makes our method the first cloud-based method to achieve maximum accuracy. The dice score is calculated by using Sørensen-Dice similarity coefficient. We also performed an extensive literature review of the brain tumor segmentation methods implemented in the last five years to get a state-of-the-art picture of well-known methodologies with a higher dice score. In comparison to the already implemented architectures, our method ranks on top in terms of accuracy in using a cloud-based 3D U-Net framework for glioma segmentation. 

/pdf/a-state-of-the-art-technique-to-perform-cloud-based-semantic-1zzvpxvc.pdf

A state-of-the-art technique to perform cloud-based semantic segmentation using deep learning 3D U-Net architecture

/pdf/a-state-of-the-art-technique-to-perform-cloud-based-semantic-e7frgndx.pdf

Two dimensional images of a person implicitly contain several useful biometric information such as gender, iris colour, weight, etc. Among them, body weight is a useful metric for a number of usecases such as forensics, fitness and health analysis, airport dynamic luggage allowance, etc. Most current solutions for body weight estimation from images make use of additional apparatus like depth sensors and thermal cameras along with predefined features such as gender and height which generally make them more computationally intensive. Motivated by the need to provide a time and cost efficient solution, a novel computer-vision based method for body weight estimation using only 2D images of people is proposed. Considering the anthropometric features from the two most common types of images, facial and full body, facial landmark measurements and body joint measurements are used in deep learning and XG boost regression models to estimate the person’s body weight. The results obtained, though comparable to previous approaches, perform much faster due to the reduced complexities of the proposed models, with facial models performing better than full body models.

/pdf/body-weight-estimation-using-2d-body-image-3aq3pxjc4u.pdf

Body Weight Estimation using 2D Body Image

Body Mass Index (BMI) is the most commonly used tool to evaluate an individual's health. It is used to classify a person as underweight, healthy weight, overweight or obese. BMI is co-related with body fat and is a vital indicator of possible diseases that can transpire with higher body fat ranges. Higher body fat is prevalent these days with a higher calorie diet and a low physical activity lifestyle. On the other end of the spectrum, Adult malnutrition is more common and widespread than we are conscious of these days. The BMI can be used as a measure of adult nutritional status, both of individuals and of communities. Given that people have less time in their busy life and most people dont own a weighing machine and/or a measuring tape, we propose a time and cost efficient method of estimating Height, Weight and BMI from a persons face. In this paper, we propose a novel model using Convolution Neural Networks (CNN) and Artificial Neural Networks (ANN). We start by detecting the face from an image using the Viola-Jones algorithm. The image is fed to the Feature Extractor model. The extracted features are passed to an Artificial Neural Network (ANN) model which gives the predicted Height, Weight and BMI values. We have evaluated our model on the Reddit-HWBMI dataset and Face-to-BMI dataset. We propose a novel dataset, the Reddit-HWBMI dataset which contains 982 subjects with their corresponding Height, Weight, BMI, Gender and Age. The best performance for BMI was given by the XceptionNet model when used as a Feature Extractor. The XceptionNet also performed best for weight, whereas VGG-Face (Resnet model) performed slightly better than XceptionNet for height.

A novel method to estimate Height, Weight and Body Mass Index from face images

Gliomas are the most hostile among brain tumors and other known cranial ailments. Patient with high-grade Gliomas often results in diminished life span. In an attempt to provide a solution, we propose a brain segmentation model for tumor detection which could be considered an appropriate tool for medical image processing. Due to the presence of an enormous quantity of magnetic resonance scanned images, performing cancer diagnosis by lab scanned images becomes quite laborious and hand-operated segmentation for brain tumor becomes complicated and slow-moving chore. One reason for its complexity could be explained by the fact that, these tumors have high probabilities of materializing in any part of the human brain without having defined size or shape. The aforementioned problems fueled our research to built a novel approach using deep learning which utilizes both global as well as local features of medical imaging for precise segmentation. In our research, we have implemented and tested Convolutional Neural Network (CNN) based models for detecting brain tumors by utilizing the concept of deep neural networks. Our proposed deep learning model combines Two-Pathway and Cascade architectures to analyse and implement brain segmentation. These models are used for segmenting Gliomas of both types. Our results are evaluated over Input Cascade and outcomes showed better performance than existing MFC Cascade.

Deep Learning Model for Brain Tumor Segmentation & Analysis

It is claimed that convolutional neural networks are inspired by human vision systems. Based on the literature of development of human visual system, we know that newly born child has blurred vision initially due to rapid eye movements. This rapid eye movement is termed as Nystagmus. This paper is concerned with a novel approach to quantify the nystagmus and implementing an artificial system that can mimic the visual learning of a newly born child or person with nystagmus. To quantify the nystagmus, we have recorded 10 seconds of eye movement videos from 3 subjects and 10 trials. We estimate the eye movement frequency by tracking the eye pupil through image processing, which is then used to create a database. To simulate a suitable learning environment, we have trained our model on gradually decreasing blurriness on Dog vs Cat dataset for classification task. The novelty of the paper is in the type of training which is elicited by human visual learning system. The results show that the performance of the network improves significantly when trained iteratively with increasing level of blur.

Ankur Haritosh

Papers

A novel method to estimate Height, Weight and Body Mass Index from face images

Deep Learning Model for Brain Tumor Segmentation & Analysis

Human Visual Learning Inspired Effective Training Methods