Breast cancer classification by a new approach to assessing deep neural network-based uncertainty quantification methods

A hybrid explainable federated-based vision transformer framework for breast cancer prediction via risk factors

Breast cancer remains a leading cause of mortality in women, underscoring the need for timely and accurate diagnosis. This paper addresses this challenge by introducing a comprehensive explainable federated learning framework for breast cancer prediction. We evaluate three deep learning approaches in both centralized and federated scenario settings: (1) individual artificial intelligence (AI) models, (2) high-level feature space ensemble models, and (3) a hybrid model combining global Vision Transformer (ViT) and local convolutional neural network (CNN) features. These models are assessed on binary, multi-class, and Breast Imaging Reporting and Data System (BI-RADS) classification tasks using a unique dataset encompassing real-world risk factors. In the federated scenario, we employ three clients with the same approaches as the centralized setting, aggregating their predictions using an AI global model. Explainable AI (XAI) technique is incorporated to enhance AI models' transparency. Our federated learning approach demonstrates superior performance, achieving accuracies of 98.65%, 97.30%, and 95.59% for binary, multi-class, and BI-RADS tasks, respectively. The proposed model, evaluated with a 95% Confidence Interval (CI) and Areas Under Curve (AUC) curves, registers top classifiers with an AUC of 0.970 [0.917-1]. Local Interpretable Model-Agnostic Explanations (LIME) XAI-based federated learning framework offers a promising solution for privacy-preserving and accurate breast cancer prediction in both research and clinical practice.

Hybrid convolutional neural network and bi-LSTM model with EfficientNet-B0 for high-accuracy breast cancer detection and classification

Breast cancer detection remains one of the most challenging problems in medical imaging. We propose a novel hybrid model that integrates Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory (Bi-LSTM) networks, and EfficientNet-B0, a pre-trained model. By leveraging EfficientNet-B0, which has been trained on the large and diverse ImageNet dataset, our approach benefits from transfer learning, enabling more efficient feature extraction from mammographic images compared to traditional methods that require CNNs to be trained from scratch. The model further enhances performance by incorporating Bi-LSTM, which allows for processing temporal dependencies in the data, which is crucial for accurately detecting complex patterns in breast cancer images. We fine-tuned the model using the Adam optimizer to optimize performance, significantly improving accuracy and processing speed. Extensive evaluation of well-established datasets such as CBIS-DDSM and MIAS resulted in an outstanding 99.2% accuracy in distinguishing between benign and malignant tumors. We also compared our hybrid model to other well-known architectures, including VGG-16, ResNet-50, and DenseNet169, using three optimizers: Adam, RMSProp, and SGD. The Adam optimizer consistently achieved the highest accuracy and lowest loss across the training and validation phases. Additionally, feature visualization techniques were applied to enhance the model's interpretability, providing deeper insight into the decision-making process. The Proposed hybrid model sets a new standard in breast cancer detection, offering exceptional accuracy and improved transparency, making it a valuable tool for clinicians in the fight against breast cancer.

CICADA (UCX): A novel approach for automated breast cancer classification through aggressiveness delineation

Breast cancer remains one of the leading causes of mortality worldwide, with current classification and segmentation techniques often falling short in accurately distinguishing between benign and malignant cases. The study both emphasize the novel approach, CICADA (UCX), specifically designed for breast segmentation with a focus on delineating aggressiveness. While the title highlights segmentation, the abstract expands on this by detailing the model's effectiveness in enhancing diagnostic precision in classifying aggressive tumor characteristics. Breast cancer segmentation pertains to the delineation of malignant tissue borders in medical imaging. The objective is to precisely delineate the malignant area from healthy tissues, facilitating reliable evaluation of tumor attributes like location, size, and form. Historically, manual segmentation by radiologists has been the benchmark; however, it is labor-intensive and susceptible to fluctuation among different observers and within the same observer. With the advancement of medical imaging technologies, there is an increasing demand for automated or semi-automatic systems capable of performing segmentation with efficiency and precision. These strategies seek to minimize human error, enhance reproducibility, and expedite diagnosis, so enabling prompt treatment. A significant problem in breast cancer segmentation is the variability in tumor morphology among various patients and imaging techniques. Neoplasms exhibit considerable variability in dimensions, morphology, and density, complicating the formulation of a universal approach. Moreover, elements like breast tissue density, which might hinder tumor appearance in mammograms, further complicate segmentation. A further barrier is the necessity for extensive, meticulously annotated datasets to train and test machine learning models, as medical picture annotation is labor-intensive and demands specialized expertise. Notwithstanding these obstacles, automated breast cancer segmentation has demonstrated significant potential in clinical applications. It assists radiologists in swiftly and precisely identifying questionable areas, resulting in earlier diagnosis and enhanced patient outcomes. Automated devices can aid in treatment planning by delivering accurate measures of tumor size and location, which are essential for establishing suitable surgical or radiation methods. This study addresses these limitations by introducing CICADA (UCX), which aims to enhance diagnostic precision and operational efficiency in clinical applications. The present study focuses on the creation and assessment of a sophisticated medical picture segmentation model, called Cheetah Inspired Convex Adaptive Discriminator Algorithm with Unet Convenet Xt CICADA (UCX), by contrasting it with the most advanced techniques currently in use. With a mean IOU of 96.34 %, a Dice Coefficient/F1-Score of 99.6461 %, and an AUC of 99.88 %, the suggested model performs quite well. The study incorporates various feature selection techniques like Particle Swarm Optimisation, Dragon Fly, Grey Wolf and our proposed novel technique named as CICADA (UCX). Through a thorough comparison analysis using many approaches, the paper highlights the advantages of CICADA (UCX) for medical picture segmentation. The study advances the area by offering fresh perspectives on segmentation accuracy, with a focus on obtaining a high Dice Coefficient/F1-Score. The results highlight how CICADA (UCX) has the ability to greatly improve medical image analysis and enable more precise and effective diagnosis. The CICADA (UCX) model, a revolutionary approach to medical picture segmentation, is presented in this study, which is a significant improvement over other existing technique. The model outperforms state-of-the-art methods in a thorough comparison investigation, showing higher performance across important assessment measures including mean IOU, Dice Coefficient/F1-Score, and AUC. Notably, the model scores a remarkable 99.6461 % Dice Coefficient/F1-Score, demonstrating accurate medical structural delineation. An important aspect of medical imaging applications is segmentation accuracy, which is greatly improved by this study. The results point to possible improvements in operational efficiency and diagnostic accuracy, which would be beneficial to patients as well as medical personnel. This discovery has significance for improving medical picture segmentation techniques and promoting technological developments in medical imaging and computer-aided diagnosis.

Application of uncertainty quantification to artificial intelligence in healthcare: A review of last decade (2013-2023)

Uncertainty estimation in healthcare involves quantifying and understanding the inherent uncertainty or variability associated with medical predictions, diagnoses, and treatment outcomes. In this era of Artificial Intelligence (AI) models, uncertainty estimation becomes vital to ensure safe decision-making in the medical field. Therefore, this review focuses on the application of uncertainty techniques to machine and deep learning models in healthcare. A systematic literature review was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Our analysis revealed that Bayesian methods were the predominant technique for uncertainty quantification in machine learning models, with Fuzzy systems being the second most used approach. Regarding deep learning models, Bayesian methods emerged as the most prevalent approach, finding application in nearly all aspects of medical imaging. Most of the studies reported in this paper focused on medical images, highlighting the prevalent application of uncertainty quantification techniques using deep learning models compared to machine learning models. Interestingly, we observed a scarcity of studies applying uncertainty quantification to physiological signals. Thus, future research on uncertainty quantification should prioritize investigating the application of these techniques to physiological signals. Overall, our review highlights the significance of integrating uncertainty techniques in healthcare applications of machine learning and deep learning models. This can provide valuable insights and practical solutions to manage uncertainty in real-world medical data, ultimately improving the accuracy and reliability of medical diagnoses and treatment recommendations.