ChestAi

Cyclical Learning Rates (CLR’S) for Improving Training Accuracies and Lowering Computational Cost

Prediction of different lung pathologies using chest X-ray images is a challenging task requiring robust training and testing accuracies. In this article, one-class classifier (OCC) and binary classification algorithms have been tested to classify 14 different diseases (atelectasis, cardiomegaly, consolidation, effusion, edema, emphysema, fibrosis, hernia, infiltration, mass, nodule, pneumonia, pneumothorax and pleural-thickening). We have utilized 3 different neural network architectures (MobileNetV1, Alexnet, and DenseNet-121) with four different optimizers (SGD, Adam, and RMSProp) for comparing best possible accuracies. Cyclical learning rate (CLR), a tuning hyperparameters technique was found to have a faster convergence of the cost towards the minima of cost function. Here, we present a unique approach of utilizing previously trained binary classification models with a learning rate decay technique for re-training models using CLR’s. Doing so, we found significant improvement in training accuracies for each of the selected conditions. Thus, utilizing CLR’s in callback functions seems a promising strategy for image classification problems.

Machine learning for identification and characterization of molecular gene signatures in progression of benign tumors

Kidney cancer is one of the most common cancer with a 2.02% risk in men and 1.02% in women. While several attempts are being made in improving early diagnosis, artificial intelligence has been successful in prediction of cancer progression. This paper utilizes artificial intelligence to classify genome-wide CpG regions for the identification of novel genes in the pathogenesis of benign and malignant cancer. Briefly, random forest is utilized for narrowing down the number of genes and later classify them with the aim of generating a diagnostic and prognostic biomarker signature. Characterizing molecular pathophysiology of identified genes with linking the methylation changes and its causative effects in renal cysts and masses development and progression is important for validation of proposed technique.

Single shot detector application for image disease localization

Bounding box algorithms are useful in localization of image patterns. Recently, utilization of convolutional neural networks on X-ray images has proven a promising disease prediction technique. However, pattern localization over prediction has always been a challenging task with inconsistent coordinates, sizes, resolution and capture positions of an image. Several model architectures like Fast R-CNN, Faster R-CNN, Histogram of Oriented Gradients (HOG), You only look once (YOLO), Region-based Convolutional Neural Networks (R-CNN), Region-based Fully Convolutional Networks (R-FCN), Single Shot Detector (SSD), etc. are used for object detection and localization in modern-day computer vision applications. SSD and region-based detectors like Fast R-CNN or Faster R-CNN are very similar in design and implementation, but SSD have shown to work efficiently with larger frames per second (FPS) and lower resolution images. In this article, we present a unique approach of SSD with a VGG-16 network as a backbone for feature detection of bounding box algorithm to predict the location of an anomaly within chest X-ray image.

K-fold Semi-supervised Self-learning Technique for Image Disease Localization

The unlabeled data can be annotated with the help of semi-supervised learning (SSL) algorithms like self-learning SSL algorithms, graph-based SSL algorithms, or the low-density separations. In this article, we have applied a self-learning semi-supervised technique with k-fold iterative training mechanism to localize cardiomegaly from the chest X-ray images. The purpose of using the iterative training method instead of the traditional self-learning semi-supervised algorithm was to generate more labeled images, increase the robustness of the algorithm and induce faster learning. The model was trained over a single-shot detector bounding box algorithm with a VGG-16 backbone for predicting the bounding box coordinates. This method generates a significant improvement in validation loss with an increase in the labeled dataset size by around 42 times compared to the traditionally used self-learning semi-supervised technique trained for the same number of epochs.

Neural Networks for Predicting Severity of Ovarian Carcinomas

Among gynecological cancers, ovarian cancer is the greatest cause of death. The lack of efficient early detection measures is linked to diagnosis at an advanced stage and a bad prognosis. Several genes have been shown to have significantly expressed in early and late stages of ovarian cancer. The use of a neural network can help infer meaning and find patterns from large data sets. The benefit of a neural network is that it is adaptive in nature, learning from the information it receives and adjusting its weights for a better prediction in instances when the outcome is unknown. Some studies have shown effective utilization of neural networks with miRNA data in ovarian cancers, others have used it in gastric cancers with survival data sets, and this study focuses on testing them on gene expression data in ovarian cancers. We found a robust RMSE values for prediction, a reasonable k-fold cross validation, and robust cross grade-stage predictions. This study defines a clear scope of utilizing neural networks in predicting grades and stages in ovarian cancers.

Stroke Risk Stratification Using Neural Networks

Current primary stroke preventive strategies seem insufficient in light of the increased prevalence of stroke, the steady or increasing death rates from cardiovascular illnesses, and the expanding list of stroke risk factors. A class of computer algorithms known as machine learning (ML) can learn from data without having to be explicitly programmed. To predict stroke and its effects, a number of physiological and clinical indicators have been used. A cyber-physical stroke rehabilitation system (CP-SRS) as well as the modified Rankin Scale (mRS90) and National Institutes of Health Stroke Scale (NIHSS24) have both been predicted using ANN models. The results of this study indicate that neural networks might create a new and efficient way to categorize stroke patients’ risk.

Addressing Class Imbalance Problem in Semantic Segmentation Using Binary Focal Loss

Image segmentation is a foundational technique in computer vision with wide-ranging applications, including its critical role in medical imaging for object identification, automatic labeling, and disease diagnosis. Advancements in deep learning have significantly improved the accuracy and efficiency of image segmentation, making it an increasingly valuable tool in various domains. Class imbalanced datasets are a frequent problem experienced when trying to train segmentation networks. Class imbalance occurs when some classes (semantic categories) in the image have significantly more instances (pixels) than others. In semantic segmentation, this often happens because certain object categories are more prevalent in the real world or dataset, while others are rarer. When training a deep learning model for semantic segmentation, this imbalance can lead to several problems. In this article, we have experimented with the class weightage parameters of binary focal loss to address the class imbalance problem in semantic segmentation. By utilizing the CANDID-PTX dataset, we have utilized U-Net architecture containing upsampling (encoder) and a downsampling (decoder) network for comparing binary focal loss rates among different alpha and gamma coefficients class weights. Doing so, we found that the adjustment of class weights in the loss function could notably help in resolving the class imbalance problems.

Common cancer biomarkers of breast and ovarian types identified through artificial intelligence

Biomarkers can offer great promise for improving prevention and treatment of complex diseases such as cancer, cardiovascular diseases, and diabetes. These can be used as either diagnostic or predictive or as prognostic biomarkers. The revolution brought about in biological big data analytics by artificial intelligence (AI) has the potential to identify a broader range of genetic differences and support the generation of more robust biomarkers in medicine. AI is invigorating biomarker research on various fronts, right from the cataloguing of key mutations driving the complex diseases like cancer to the elucidation of molecular networks underlying diseases. In this study, we have explored the potential of AI through machine learning approaches to propose that these methods can act as recommendation systems to sort and prioritize important genes and finally predict the presence of specific biomarkers. Essentially, we have utilized microarray datasets from open-source databases, like GEO, for breast, lung, colon, and ovarian cancer. In this context, different clustering analyses like hierarchical and k-means along with random forest algorithm have been utilized to classify important genes from a pool of several thousand genes. To this end, network centrality and pathway analysis have been implemented to identify the most potential target as CREB1.

OpenAI's GPT-3/ChatGPT: Suggests & Correct

OpenAI's ChatGPT is a GPT-3-based chatbot that has been optimized through supervised and reinforcement learning methodologies. The training data includes information about internet phenomena and programming languages like Python. However, ChatGPT has limitations, such as over-optimized reward models and limited training data from 2021. This study tests the effectiveness of ChatGPT in syntax auto-correction and output generation from Python code libraries.

Evaluating Differences in Small Object Localization Using Semantic Segmentation and Single Shot Detector (SSD) Bounding Box Algorithm

In the computer vision task of semantic segmentation, each pixel in an image is classified into a particular class or category. Unlike object detection, which detects objects and provides bounding boxes, semantic segmentation assigns a class label to each individual pixel. This results in a pixel-wise classification map that identifies different objects or regions within an image. In the present study, we have tried to compare two very popular techniques used for detecting the object in an image using computer vision. Semantic segmentation and bounding box algorithms have been compared in a controlled environment, keeping the dataset, train-test split, batch size, training epochs, and other factors constant. We have compared these 2 techniques for loss minimization and area of overlap functions. A gradual decrease in the training loss has been observed in semantic segmentation technique, while the bounding box algorithms generate a steep decrease specifically in the later epochs. Choosing appropriate object detection techniques should address the problem of small regions of interest (ROI) compared to the total image area of class imbalance problem in semantic segmentation.