Deep Learning Approach for Medical Image Analysis

CosSIF: Cosine similarity-based image filtering to overcome low inter-class variation in synthetic medical image datasets

Crafting effective deep learning models for medical image analysis is a complex task, particularly in cases where the medical image dataset lacks significant inter-class variation. This challenge is further aggravated when employing such datasets to generate synthetic images using generative adversarial networks (GANs), as the output of GANs heavily relies on the input data. In this research, we propose a novel filtering algorithm called Cosine Similarity-based Image Filtering (CosSIF). We leverage CosSIF to develop two distinct filtering methods: Filtering Before GAN Training (FBGT) and Filtering After GAN Training (FAGT). FBGT involves the removal of real images that exhibit similarities to images of other classes before utilizing them as the training dataset for a GAN. On the other hand, FAGT focuses on eliminating synthetic images with less discriminative features compared to real images used for training the GAN. The experimental results reveal that the utilization of either the FAGT or FBGT method reduces low inter-class variation in clinical image classification datasets and enables GANs to generate synthetic images with greater discriminative features. Moreover, modern transformer and convolutional-based models, trained with datasets that utilize these filtering methods, lead to less bias toward the majority class, more accurate predictions of samples in the minority class, and overall better generalization capabilities. Code and implementation details are available at: https://github.com/mominul-ssv/cossif.

Enhancing Skin Cancer Detection and Classification in Dermoscopic Images through Concatenated MobileNetV2 and Xception Models

One of the most promising research initiatives in the healthcare field is focused on the rising incidence of skin cancer worldwide and improving early discovery methods for the disease. The most significant factor in the fatalities caused by skin cancer is the late identification of the disease. The likelihood of human survival may be significantly improved by performing an early diagnosis followed by appropriate therapy. It is not a simple process to extract the elements from the photographs of the tumors that may be used for the prospective identification of skin cancer. Several deep learning models are widely used to extract efficient features for a skin cancer diagnosis; nevertheless, the literature demonstrates that there is still room for additional improvements in various performance metrics. This study proposes a hybrid deep convolutional neural network architecture for identifying skin cancer by adding two main heuristics. These include Xception and MobileNetV2 models. Data augmentation was introduced to balance the dataset, and the transfer learning technique was utilized to resolve the challenges of the absence of labeled datasets. It has been detected that the suggested method of employing Xception in conjunction with MobileNetV2 attains the most excellent performance, particularly concerning the dataset that was evaluated: specifically, it produced 97.56% accuracy, 97.00% area under the curve, 100% sensitivity, 93.33% precision, 96.55% F1 score, and 0.0370 false favorable rates. This research has implications for clinical practice and public health, offering a valuable tool for dermatologists and healthcare professionals in their fight against skin cancer.

StynMedGAN: Medical images augmentation using a new GAN model for improved diagnosis of diseases

Deep networks require a considerable amount of training data otherwise these networks generalize poorly. Data Augmentation techniques help the network generalize better by providing more variety in the training data. Standard data augmentation techniques such as flipping, and scaling, produce new data that is a modified version of the original data. Generative Adversarial networks (GANs) have been designed to generate new data that can be exploited. In this paper, we propose a new GAN model, named StynMedGAN for synthetically generating medical images to improve the performance of classification models.  StynMedGAN builds upon the state-of-the-art styleGANv2 that has produced remarkable results generating all kinds of natural images. We introduce a regularization term that is a normalized loss factor in the existing discriminator loss of styleGANv2. It is used to force the generator to produce normalized images and penalize it if it fails. Medical imaging modalities, such as X-Rays, CT-Scans, and MRIs are different in nature, we show that the proposed GAN extends the capacity of styleGANv2 to handle medical images in a better way. This new GAN model (StynMedGAN) is applied to three types of medical imaging: X-Rays, CT scans, and MRI to produce more data for the classification tasks. To validate the effectiveness of the proposed model for the classification, 3 classifiers (CNN, DenseNet121, and VGG-16) are used. Results show that the classifiers trained with StynMedGAN-augmented data outperform other methods that only used the original data. The proposed model achieved 100%, 99.6%, and 100% for chest X-Ray, Chest CT-Scans, and Brain MRI respectively. The results are promising and favor a potentially important resource that can be used by practitioners and radiologists to diagnose different diseases.

MobileNet-SVM: A Lightweight Deep Transfer Learning Model to Diagnose BCH Scans for IoMT-Based Imaging Sensors

Many individuals worldwide pass away as a result of inadequate procedures for prompt illness identification and subsequent treatment. A valuable life can be saved or at least extended with the early identification of serious illnesses, such as various cancers and other life-threatening conditions. The development of the Internet of Medical Things (IoMT) has made it possible for healthcare technology to offer the general public efficient medical services and make a significant contribution to patients' recoveries. By using IoMT to diagnose and examine BreakHis v1 400× breast cancer histology (BCH) scans, disorders may be quickly identified and appropriate treatment can be given to a patient. Imaging equipment having the capability of auto-analyzing acquired pictures can be used to achieve this. However, the majority of deep learning (DL)-based image classification approaches are of a large number of parameters and unsuitable for application in IoMT-centered imaging sensors. The goal of this study is to create a lightweight deep transfer learning (DTL) model suited for BCH scan examination and has a good level of accuracy. In this study, a lightweight DTL-based model "MobileNet-SVM", which is the hybridization of MobileNet and Support Vector Machine (SVM), for auto-classifying BreakHis v1 400× BCH images is presented. When tested against a real dataset of BreakHis v1 400× BCH images, the suggested technique achieved a training accuracy of 100% on the training dataset. It also obtained an accuracy of 91% and an F1-score of 91.35 on the test dataset. Considering how complicated BCH scans are, the findings are encouraging. The MobileNet-SVM model is ideal for IoMT imaging equipment in addition to having a high degree of precision. According to the simulation findings, the suggested model requires a small computation speed and time.

Deep Multi-Scale Residual Connected Neural Network Model for Intelligent Athlete Balance Control Ability Evaluation

Athlete balance control ability plays an important role in different types of sports. Accurate and efficient evaluations of the balance control abilities can significantly improve the athlete management performance. With the rapid development of the athlete training field, intelligent and automatic evaluations have been highly demanded in the past years. This study proposes a deep learning-based athlete balance control ability evaluation method through processing the time-series movement pressure measurement data. An end-to-end model structure is proposed, which directly analyzes the raw data and provides the evaluation results, which largely facilitates practical utilization. A multi-scale feature extraction scheme is employed, by exploring the learned features in different scales. A residual connected neural network architecture is further proposed. By using the short-cut connection, the deep neural network model can be more efficiently trained. Experiments on the real athlete balance control ability tests are carried out for validations. Through comparisons with different related methods, the results show the proposed deep multi-scale residual connected neural network model is well suited for the athlete balance control ability evaluation problem, and promising for actual applications in the real scenarios.

CAttSleepNet: Automatic End-to-End Sleep Staging Using Attention-Based Deep Neural Networks on Single-Channel EEG

Accurate sleep staging results can be used to measure sleep quality, providing a reliable basis for the prevention and diagnosis of sleep-related diseases. The key to sleep staging is the feature representation of EEG signals. Existing approaches rarely consider local features in feature extraction, and fail to distinguish the importance of critical and non-critical local features. We propose an innovative model for automatic sleep staging with single-channel EEG, named CAttSleepNet. We add an attention module to the convolutional neural network (CNN) that can learn the weights of local sequences of EEG signals by exploiting intra-epoch contextual information. Then, a two-layer bidirectional-Long Short-Term Memory (Bi-LSTM) is used to encode the global correlations of successive epochs. Therefore, the feature representations of EEG signals are enhanced by both local and global context correlation. Experimental results achieved on two real-world sleep datasets indicate that the CAttSleepNet model outperforms existing models. Moreover, ablation experiments demonstrate the validity of our proposed attention module.

Neuronal Apoptosis in Patients with Liver Cirrhosis and Neuronal Epileptiform Discharge Model Based upon Multi-Modal Fusion Deep Learning

Neurons refer to nerve cells. Each neuron is connected with thousands of other neurons to form a corresponding functional area and carry out complex communication with other functional areas. Its importance to the human body is self-evident. There are also many scholars studying the mechanism of apoptosis. This paper proposes a study of neuronal apoptosis in patients with liver cirrhosis and neuronal epileptiform discharge models based on multi-modal fusion deep learning, aiming to study the influencing factors of abnormal neuronal discharge in the brain. The method in this paper is to study multi-modal information fusion methods, perform Bayesian inference, and analyze multi-modal medical data. The function of these research methods is to obtain the relationship between the independence of information and the intersection of information among modalities. In the neuronal epileptiform discharge model, the mRNA expression level of the necroptotic signaling pathway related protein was detected, and the mechanism of neuronal necrosis in patients with liver cirrhosis was explored. Experiments show that the neuron recognition rate has been increased from 67.2% to 84.5%, and the time has been reduced, proving the effectiveness of deep learning.

Finite Element Analysis of Femoral-Acetabular Impingement (FAI) Based on Three-Dimensional Reconstruction

In order to solve the problem that people often have pain in the hip joint, it is more meaningful to study femoral-acetabular impingement syndrome in the future. This article aims to study the finite element analysis of femoral-acetabular impingement based on three-dimensional reconstruction. This paper proposes a selective image matching strategy. In the feature matching stage, all images are not matched in pairs, but the corresponding camera distance between the images is calculated initially, which has little effect on the number of features and greatly reduces the time of feature matching, thereby reducing the time cost of 3D reconstruction. In this experiment, a double-blind experiment was used to check the range of motion of all hip joints. Two senior radiologists read the obtained hip joint orthographic films to screen out the hip joint orthographic films that meet the requirements. Experimental data shows that although the initial matching points of the algorithm in this paper are lower than those of the traditional algorithm, the final number of matching points is higher than that of the traditional algorithm. When the final number of patches is fixed to 10000, the initial patch required by the algorithm in this paper is more than that required by the SAD algorithm, nearly 13%, but the total storage requirement is 56.4% of the SAD algorithm, which is a big improvement.

Attentive 3D-Ghost Module for Dynamic Hand Gesture Recognition with Positive Knowledge Transfer

Hand gesture recognition is a challenging topic in the field of computer vision. Multimodal hand gesture recognition based on RGB-D is with higher accuracy than that of only RGB or depth. It is not difficult to conclude that the gain originates from the complementary information existing in the two modalities. However, in reality, multimodal data are not always easy to acquire simultaneously, while unimodal RGB or depth hand gesture data are more general. Therefore, one hand gesture system is expected, in which only unimordal RGB or Depth data is supported for testing, while multimodal RGB-D data is available for training so as to attain the complementary information. Fortunately, a kind of method via multimodal training and unimodal testing has been proposed. However, unimodal feature representation and cross-modality transfer still need to be further improved. To this end, this paper proposes a new 3D-Ghost and Spatial Attention Inflated 3D ConvNet (3DGSAI) to extract high-quality features for each modality. The baseline of 3DGSAI network is Inflated 3D ConvNet (I3D), and two main improvements are proposed. One is 3D-Ghost module, and the other is the spatial attention mechanism. The 3D-Ghost module can extract richer features for hand gesture representation, and the spatial attention mechanism makes the network pay more attention to hand region. This paper also proposes an adaptive parameter for positive knowledge transfer, which ensures that the transfer always occurs from the strong modality network to the weak one. Extensive experiments on SKIG, VIVA, and NVGesture datasets demonstrate that our method is competitive with the state of the art. Especially, the performance of our method reaches 97.87% on the SKIG dataset using only RGB, which is the current best result.