Among women, breast cancer remains one of the most dominant cancer types. In the year 2022, around 2,87,800 new cases were diagnosed, and 43,200 women faced mortality due to this disease. Analysis and processing of mammogram images is vital for its earlier identification and thus helps in reducing mortality rates and facilitating effective treatment for women. Accordingly, several deep-learning techniques have emerged for mammogram classification. However, it is still challenging and requires promising solutions. This study proposed a newer automated computer-aided implementation for breast cancer classification. The work starts with enhancing the mammogram contrast using a haze-reduced adaptive technique followed by augmentation. Afterward, EfficientNet-B4 pre-trained architecture is trained for both original and enhanced sets of mammograms individually using static hyperparameters’ initialization. This provides an output of 1792 feature vectors for each set and then fused using a serial mid-value-based approach. The final feature vectors are then optimized using a chaotic-crow-search optimization algorithm. Finally, the obtained significant feature vectors were classified with the aid of machine learning algorithms. The evaluation is made using INbreast and CBIS-DDSM databases. The proposed framework attained a balanced computation time with a maximum classification performance of 98.459 and 96.175% accuracies on INbreast and CBIS-DDSM databases, respectively.

Medical IoT devices and Telemedicine computation is the growing domain and further involving biomedical computation via machine learning ecosystem has generated an insightful results and analysis. The resources sharing and availability in computing and decision support suffer with a higher latency and energy consumption. In this manuscript, a novel TinyML based model for medical consumer devices resources allocation and resource sharing is discussed. The proposed framework is developed using Federated learning (FL) models for extracting the resource utilization patterns at individual user levels. These locally computed models are further facilitated with edge computation layer for locating resource patterns extraction. The technique is deployed on the dynamic server based resource pooling for effective analysis and resource scheduling and expanded to develop a reliable recommendation model for medical resource management. The framework has trained 128 clusters of 6400 rural and 12800 urban IoT devices samples for resource allocation and scheduling using telemedicine protocol (TelMED). The framework has secured an efficiency of 93.21% in urban user recommendation and 94.72% for rural users.

The Internet of Things (IoT) and associated capabilities are becoming indispensable in the planning, operation, and administration of intricate systems of all sizes. High-end learning solutions that go beyond the boundaries of the problem are necessary for addressing the variety of communication concerns (compatibility, secure communication, etc.) in IoT settings. Building machine learning (ML) networks from disparate data sources is a cutting-edge practice known as Federated Learning (FL). In this article, we implement FL between edge-based servers and devices in a sparsely populated cloud to facilitate cohesive learning and the storage of critical information in smart IoT systems. FL enables collaborative training from a common model by aggregating smaller unit models via regulated edge network participants. Further, all the susceptible device’s information and sensitive message transactions are addressed via blockchain technology. Thus, a blockchain-based security mechanism is integrated to secure user privacy and facilitate widespread practical adoption. Finally, a comparison is made between the proposed model and the three best free, open-source Federated Learning models already in use (FedPD, FedProx, and FedAvg). In terms of statistical, and data heterogeneity (>70% SDI, >97% accuracy), the experimental findings suggest that the proposed model performs better than the existing techniques.

Chronic kidney disease (CKD) affects millions of people worldwide, greatly reducing their quality of life and creating serious economic, social, and medical problems. Some automated diagnosis methods can detect chronic renal disease. In-depth studies on data mining techniques have recently focused on accuracy in the diagnosis of chronic renal illnesses, either by taking advantage of the disease's simplicity or doing feature selection in addition to pre-processing. In order to handle the unbalanced dataset in this work, Synthetic Minority Over Sampling Technique (SMOTE) is used during pre-processing. For this investigation, 400 data from the publicly accessible UCI machine learning (ML) repository are used. For the implementation, both homogeneous and heterogeneous ensemble classifiers which combine two separate classifiers have been used. Different machine learning (ML) techniques, such as the Classification and Regression Tree (CART), Adaboost classifier, Decision Tree (DT), Reduced Error Pruning Tree, Alternating Decision Tree, and Random Forests Algorithm and their ensembles with a significant reduction in entropy, are used to perform the classification. With a 99.12% accuracy rate and a 99.10% f1 score, the homogeneous classifier Adaboost-Random Forest outperforms other models in the prediction of CKD.

Privacy in medical records while data sharing is a major concern for distributed learning models. The dataset generated and shared via Electronic Medical Records (EMR) consist of sensitive medical information such as patient identify and experts recommendations, and causes setbacks in training larger models, dataset augmentation and polluting datasets with recursive attributes. The information processing and de-identification is proposed in this article to preserve and enhance the privacy of EMR. The proposed technique is termed as PrEGAN (i.e.) Privacy Enhanced Generative Adversarial Network (GAN) for EMR data training and realistic mapping. The proposed model generates and discriminates the ground truth with generated mask via a computation of loss function for de-identification or removal of personal linked/connected data in the records networks. The objective is to generate the mask of EMR, which is realistic and similar to the ground truth. The model is trained and validated with two distinguished discriminators, the CNN based discriminator is used for medical images, whereas Neural Networks are used for textural data generator. The experimental results demonstrate a higher degree of data privacy and de-identification in EMR with 88.32% accuracy in predicting and eliminating via RoI and loss function.