Mohammad Farukh Hashmi

1. Peripheral capillary oxygen saturation (SpO2) indicates the percentage of oxygen-bound hemoglobin in arterial blood and serves as a vital clinical parameter for assessing respiratory and circula...

Digital video content must be stored and transmitted with efficient video compression, particularly in environments with limited bandwidth, such as mobile communication networks and internet streaming services. The two primary challenges with traditional video compression methods are minimizing latency and maintaining video quality while achieving high compression ratios. This paper presents a novel framework that combines a temporal-context-adaptive entropy model, recurrent auto-encoders (RAE), and recurrent probability models (RPM) to address these issues. The main goal of the framework is to efficiently encode latent representations and record temporal correlations between consecutive video frames in order to achieve high compression rates. The RAE uses Convolutional Long Short-Term Memory (ConvLSTM) cells in its encoder and decoder networks, which leverage temporal dependencies to enhance compression. Furthermore, the suggested model uses Feature Transformation (FT) layers to integrate multi-scale motion information, which enhances compression efficiency. Adaptive arithmetic coding that is contextually informed by the temporal relationships of video frames is made possible by the RPM’s recurrent estimation of parameters of conditional probability mass functions. This technique produces more effective compression by reducing the size of latent representations while maintaining crucial video data. Implemented on the Python platform, the proposed model achievesa Peak Signal-to-Noise Ratio(PSNR) of 48.42, out performing existing compression techniques in both efficiency and quality.

Pneumonia is a major cause of death worldwide, with nearly 2.2 million deaths every year, including over 700,000 children aged five years and below. Chest X-ray (CXR) imaging is the standard method taken up to diagnose pneumonia; however, such image inspection is shown to be difficult even for expert radiologists. The intricacy in the visual patterns generated in X-ray images often results in misdiagnosis, indicating the importance of efficient and accurate automated substitutes. In this paper, we present a new machine learning-based system that incorporates deep learning combined with handcrafted feature extraction techniques for sophisticated pneumonia detection. We use ResNet-50 for deep feature extraction, with the integration of 2D-discrete wavelet transform (DWT) and Gray-Level Co-occurrence Matrix (GLCM) for texture feature extraction, for the intention of gaining helpful spatial and frequency-domain features. The features obtained are inputted into a Support Vector Machine (SVM) classifier, which is optimized for high accuracy and robust prediction. The experimental findings indicate that the proposed model produces a classification accuracy of 97%, accompanied by an F1-score of 0.97, over traditional methods. Through the synergistic integration of handcrafted and deep learning-based feature extraction techniques, our approach presents a trustworthy and efficient solution for automated pneumonia detection. The proposed method has the potential to aid radiologists in providing timely and accurate diagnoses, thus enhancing patient outcomes and curtailing the global burden of pneumonia-related mortality.

Segmentation plays a key role in designing a CAD system. Therefore, we need to implement highly accurate image segmentation to extract the shape of the lung from X-rays. Lung segmentation is challenging due to significant shape variations and unclear lung regions caused by severe lung diseases. Many CNN models give better results in segmentation, but every model has its limitations. Our paper proposes a new, improved model that builds upon the base model, SegNet. SegNet is the best model for semantic segmentation. Our proposed model is a combination of SegNet and ResNet 50. We are replacing the encoder of the SegNet with ResNet50. We modified the base model to get better performance and results. We utilized the Shenzhen dataset, a publicly available dataset, for training and testing the model. The model performance is evaluated in terms of global accuracy, Jaccard index, and dice similarity coefficient. The proposed model achieved a global accuracy of 97.73%, a Jaccard index of 97.32%, a Dice similarity coefficient of 97.33%, a precision of 98.74%, a recall of 98.58%, and an F1-score of 97.37%. The experimental results showed that our model performed better than the other models in terms of global accuracy, Jaccard index, precision, recall, F1-score, and dice similarity coefficient.

Skin cancer, particularly melanoma, arises from DNA damage that leads to abnormal cell growth in the epidermis. Early detection is crucial as melanoma can spread rapidly, but it is highly curable if identified promptly. Detecting and diagnosing melanoma early are essential to reduce mortality rates associated with this type of cancer. In the literature, various ensemble techniques have been proposed to improve the performance. This paper introduces a deep learning based ensemble method aimed at enhancing the accuracy of melanoma skin cancer detection. Additionally, it presents a thorough performance evaluation of five ensemble techniques. Initially, the dataset underwent pre-processing, involving the removal of artifacts through hair removal, and achieving a balance in the distribution of images for each class through image augmentation techniques. Then, the architecture of 16 pre-trained models was modified by adding additional layers to improve their performance. The models that achieved the highest melanoma accuracy were selected for ensembling. Since VGG16, MobileNetV2, and DenseNet169 achieved the highest melanoma accuracy, they were chosen for ensembling. Five ensemble techniques, namely, weighted average, voting, bagging, boosting, and stacking, were applied to the modified architectures of fine-tuned pre-trained models such as VGG16, MobileNetV2, and DenseNet169 to classify skin cancer images. The experiments were performed on a combined dataset of HAM10000 and ISIC2019, which contains images of seven skin lesion classes. The results demonstrate that the weighted average ensemble model achieves highest overall accuracy of 81.99% and melanoma classification accuracy of 89.85%. The positive outcomes affirm that employing ensemble techniques with adjusted model architectures enhances performance, thereby demonstrating their potential utility in the classification of skin cancer images.