Enhanced MRI-based brain tumor classification with a novel Pix2pix generative adversarial network augmentation framework

Brain Communications, Год журнала: 2024, Номер 6(6)

Опубликована: Янв. 1, 2024

Abstract The scarcity of medical imaging datasets and privacy concerns pose significant challenges in artificial intelligence-based disease prediction. This poses major to patient confidentiality as there are now tools capable extracting information by merely analysing patient’s data. To address this, we propose the use synthetic data generated generative adversarial networks a solution. Our study pioneers utilisation novel Pix2Pix network model, specifically ‘image-to-image translation with conditional networks,’ generate for brain tumour classification. We focus on classifying four types: glioma, meningioma, pituitary healthy. introduce deep convolutional neural architecture, developed from architectures, process pre-processed original obtained Kaggle repository. evaluation metrics demonstrate model's high performance images, achieving an accuracy 86%. Comparative analysis state-of-the-art models such Residual Network50, Visual Geometry Group 16, 19 InceptionV3 highlights superior our model detection, diagnosis findings underscore efficacy augmentation technique creating accurate classification, offering promising avenue improved prediction treatment planning.

Язык: Английский

---

Considering the IMPACT framework to understand the AI-well-being-complex from an interdisciplinary perspective

Telematics and Informatics Reports, Год журнала: 2023, Номер 13, С. 100112 - 100112

Опубликована: Дек. 27, 2023

Artificial intelligence (AI) is built into many products and has the potential to dramatically impact societies around world. This short theoretical paper aims provide a simple framework that might help us understand how introduction and/or use of with AI influence well-being humans. It proposed considering dynamic Interplay between variables stemming from Modality, Person, Area, Culture Transparency categories will on well-being. The Modality category encompasses areas such as degree being interactive, informational versus actualizing, or autonomous. Person variable contains age, gender, personality, technological self-efficacy, perceived competence when interacting AI, whereas Area can comprise certain product where in-built domain used make difference (such health sector, military education etc.). importance because cultural settings shape attitudes towards AI. Finally, this also be true for transparent (or understandable/explainable AI), high degrees transparency likely elicit trust. model suggests there no easy answer one seeks world Only by myriad number in model, summed up acronym IMPACT (Interaction Modality-Person-Area-Culture-Transparency), we get closer an understanding impacts individuals'

Язык: Английский

---

A novel approach of brain-computer interfacing (BCI) and Grad-CAM based explainable artificial intelligence: Use case scenario for smart healthcare

Journal of Neuroscience Methods, Год журнала: 2024, Номер 408, С. 110159 - 110159

Опубликована: Май 7, 2024

Язык: Английский

---

Artificial Intelligence Techniques and Pedigree Charts in Oncogenetics: Towards an Experimental Multioutput Software System for Digitization and Risk Prediction

Computation, Год журнала: 2024, Номер 12(3), С. 47 - 47

Опубликована: Март 3, 2024

Pedigree charts remain essential in oncological genetic counseling for identifying individuals with an increased risk of developing hereditary tumors. However, this valuable data source often remains confined to paper files, going unused. We propose a computer-aided detection/diagnosis system, based on machine learning and deep techniques, capable the following: (1) assisting oncologists digitizing paper-based pedigree charts, generating new digital ones, (2) automatically predicting predisposition directly from these charts. To best our knowledge, there are no similar studies current literature, consequently, utilization software artificial intelligence has been made public yet. By incorporating medical images other omics sciences, is also fertile ground training additional systems, broadening predictive capabilities. plan bridge gap between scientific advancements practical implementation by modernizing enhancing existing services. This would mark pioneering development AI-based application designed enhance various aspects counseling, leading improved patient care field oncogenetics.

Язык: Английский

---

Improved EfficientNet Architecture for Multi-Grade Brain Tumor Detection

Electronics, Год журнала: 2025, Номер 14(4), С. 710 - 710

Опубликована: Фев. 12, 2025

Accurate detection and diagnosis of brain tumors at early stages is significant for effective treatment. While numerous methods have been developed tumor classification, several rely on traditional techniques, often resulting in suboptimal performance. In contrast, AI-based deep learning techniques shown promising results, consistently achieving high accuracy across various types while maintaining model interpretability. Inspired by these advancements, this paper introduces an improved variant EfficientNet multi-grade addressing the gap between performance explainability. Our approach extends capabilities to classify four types: glioma, meningioma, pituitary tumor, non-tumor. For enhanced explainability, we incorporate gradient-weighted class activation mapping (Grad-CAM) improve The input MRI images undergo data augmentation before being passed through feature extraction phase, where underlying patterns are learned. achieves average 98.6%, surpassing other state-of-the-art standard datasets a substantially reduced parameter count. Furthermore, explainable AI (XAI) analysis demonstrates model’s ability focus relevant regions, enhancing its This accurate interpretable classification has potential significantly aid clinical decision-making neuro-oncology.

Язык: Английский

---

Next-generation agentic AI for transforming healthcare

Informatics and Health, Год журнала: 2025, Номер 2(2), С. 73 - 83

Опубликована: Апрель 8, 2025

Язык: Английский

---

A visual attention-based algorithm for brain tumor detection using an on-center saliency map and a superpixel-based framework

Healthcare Analytics, Год журнала: 2024, Номер 5, С. 100323 - 100323

Опубликована: Март 26, 2024

Brain tumors are life-threatening and typically identified by experts using imaging modalities like Magnetic Resonance Imaging (MRI), Computed Tomography (CT), Positron Emission (PET). However, any error due to human intervention in brain anomaly detection can have devastating consequences. This study proposes a tumor algorithm for MRI images. Previous research into has drawbacks, paving the way further investigations. A visual attention-based technique is proposed overcome these drawbacks. wide range of intensity, varying from inner matter-alike intensity skull-alike making them difficult threshold. Thus, unique approach threshold entropy been utilized. An on-center saliency map accurately captures biological attention-focused tumorous region original image. Later, superpixel-based framework used capture true structure tumor. Finally, it was experimentally shown that outperforms existing algorithms detection.

Язык: Английский

---

Estimation of Fractal Dimension and Segmentation of Brain Tumor with Parallel Features Aggregation Network

Fractal and Fractional, Год журнала: 2024, Номер 8(6), С. 357 - 357

Опубликована: Июнь 14, 2024

The accurate recognition of a brain tumor (BT) is crucial for diagnosis, intervention planning, and the evaluation post-intervention outcomes. Conventional methods manually identifying delineating BTs are inefficient, prone to error, time-consuming. Subjective BT biased because diffuse irregular nature BTs, along with varying enhancement patterns coexistence different components. Hence, development an automated diagnostic system vital mitigating subjective bias achieving speedy effective segmentation. Recently developed deep learning (DL)-based have replaced methods; however, these DL-based still low performance, showing room improvement, limited heterogeneous dataset analysis. Herein, we propose parallel features aggregation network (PFA-Net) robust segmentation three regions in scan, perform analysis validate its generality. (PFA) module exploits local radiomic contextual spatial at low, intermediate, high levels types tumors aggregates them fashion. To enhance capabilities proposed framework, introduced fractal dimension estimation into our system, seamlessly combined as end-to-end task gain insights complexity irregularity structures, thereby characterizing intricate morphology BTs. PFA-Net achieves Dice scores (DSs) 87.54%, 93.42%, 91.02%, enhancing region, whole core respectively, multimodal (BraTS)-2020 open database, surpassing performance existing state-of-the-art methods. Additionally, validated another database progression DS 64.58% analysis,

Язык: Английский

---

Brain magnetic resonance images segmentation via improved mixtures of factor analyzers based on dynamic co-clustering

Neurocomputing, Год журнала: 2024, Номер 583, С. 127551 - 127551

Опубликована: Март 14, 2024

Язык: Английский

---

Knowledge distillation in transformers with tripartite attention: Multiclass brain tumor detection in highly augmented MRIs

Journal of King Saud University - Computer and Information Sciences, Год журнала: 2023, Номер 36(1), С. 101907 - 101907

Опубликована: Дек. 28, 2023

The advent of attention-based architectures in medical imaging has ushered an era precision diagnostics, particularly the detection and classification brain tumors. This study introduced innovative knowledge distillation framework employing a tripartite attention mechanism within transformer encoder models, specifically tailored for identification multiple tumor classes through magnetic resonance (MRI). proposed methodology synergistically harnesses capabilities large, highly parameterized teacher models to train more compact, efficient student suitable deployment resource-constrained environments such as internet things smart healthcare devices. Utilizing diverse array MRI sequences—including T1, contrast-enhanced T2—this accounts nuanced variations across derived from three extensive datasets. addresses limitation traditional by innovatively integrating temperature-softening neighborhood attention, global cross-attention layers. sophisticated approach allows richer feature representation, capturing both local contextual information intricate features scans. is supplemented unique augmentation pipeline shifted patch tokenization technique, which enrich model's input especially underrepresented classes. Through meticulous experimentation ablation studies, demonstrates that model not only retains robustness its larger counterparts but also delivers enhanced performance metrics. When juxtaposed with benchmarking models—including deep CNNs various transformer-based architectures—the consistently showcases superior results. Its effectiveness reflected lower losses, commendable Brier scores, noteworthy top-1 top-5 accuracies, well AUC metrics all paper validates efficacy complex image analysis tasks provides promising pathway integration cutting-edge AI techniques real-world clinical applications, potentially revolutionizing early treatment

Язык: Английский

---