HViT4Lung: Hybrid Vision Transformers Augmented by Transfer Learning to Enhance Lung Cancer Diagnosis

Published: June 7, 2023

Lung cancer is the leading cause of mortality among other forms worldwide. Early and accurate recognition lung nodules crucial for better life quality patients. Although chest Computed Tomography (CT) scan principal imaging procedure to evaluate recognize cancer, radiologists evaluation based on CT images subjective afflicted from a low accuracy compared post-surgery pathological tests. Computer Aided Diagnosis (CAD) has been proven be beneficial in this context by increasing minimizing expert involvement. Nevertheless, due various factors including size location inconsistency nodules, errorless detection cancerous cases still challenge CAD systems. Motivated fact, paper presents novel effective method, called HViT4Lung (Hybrid Vision Transformers detection), enhance diagnosis. The proposed deep learning-based hybrid framework combines Convolution Neural Networks, augmented transfer learning that extracts features detect predict their malignancy. pipeline implemented with blocks tested sample dataset. results model are very promising approaches field, achieving 99.20% training accuracy, 99.09% validation testing classification scans 1190 into three different classes normal, benign, malignant.

Language: Английский

---

Quality of T2-weighted MRI re-acquisition versus deep learning GAN image reconstruction: A multi-reader study

European Journal of Radiology, Journal Year: 2023, Volume and Issue: 170, P. 111259 - 111259

Published: Dec. 12, 2023

Language: Английский

---

Innovative Financial Management in Higher Education: A Multi-Scale Deep Learning Approach for Risk Reduction and Quality Enhancement

Deleted Journal, Journal Year: 2024, Volume and Issue: 20(7s), P. 114 - 120

Published: May 4, 2024

This paper presents a novel university financial management system leveraging multi-scale deep learning. With rising college enrollment and teaching complexities, traditional models require adaptation to mitigate risks improve quality. The integrates hardware software innovations: multiple sensors enhance data scanning, coordinated by central coordinator, ensuring comprehensive database coverage. Software-wise, structured establishes attribute-based connections, crucial for weight assignment. Employing multilayer perceptual network topology, full interconnection model based on learning facilitates profound extraction. Experimental evaluations demonstrate the system's superior risk assessment capabilities compared approaches, extracting broader spectrum of parameters warnings. By embracing learning, this promises significant advancements in management, enhancing adaptability mitigation finance departments.

Language: Английский

---

Deep Learning Enhances Detection of Extracapsular Extension in Prostate Cancer from mpMRI of 1001 Patients

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: May 21, 2024

Abstract Extracapsular extension (ECE) is detected in approximately one-third of newly diagnosed prostate cancer (PCa) cases at stage T3a or higher and associated with increased rates positive surgical margins early biochemical recurrence following radical prostatectomy (RP). This study presents the development AutoRadAI, an end-to-end, user-friendly artificial intelligence (AI) pipeline designed for identification ECE PCa through analysis multiparametric MRI (mpMRI) fused histopathology. The dataset consists 1001 patients, including 510 pathology-confirmed 491 negative cases. AutoRadAI integrates comprehensive preprocessing followed by a sequence two novel deep learning (DL) algorithms within multi-convolutional neural network (multi-CNN) strategy. exhibited strong performance during its evaluation. In blind testing phase, achieved area under curve (AUC) 0.92 assessing image quality 0.88 detecting presence individual patients. Additionally, implemented as web application, making it ideally suited clinical applications. Its data-driven accuracy offers significant promise diagnostic treatment planning tool. Detailed instructions full are available https://autoradai.anvil.app on our GitHub page https://github.com/PKhosravi-CityTech/AutoRadAI .

Language: Английский

---

A Classifier Model Using Fine-Tuned Convolutional Neural Network and Transfer Learning Approaches for Prostate Cancer Detection

Applied Sciences, Journal Year: 2024, Volume and Issue: 15(1), P. 225 - 225

Published: Dec. 30, 2024

Background: Accurate and reliable classification models play a major role in clinical decision-making processes for prostate cancer (PCa) diagnosis. However, existing methods often demonstrate limited performance, particularly when applied to small datasets binary problems. Objectives: This study aims design fine-tuned deep learning (DL) model capable of classifying PCa MRI images with high accuracy evaluate its performance by comparing it various DL architectures. Methods: In this study, basic convolutional neural network (CNN) was developed subsequently optimized using techniques such as L2 regularization, Tanh activation, dropout, early stopping enhance performance. Additionally, pyramid-type CNN architecture designed simultaneously both fine details broader structures combining low- high-resolution information through feature maps extracted from different layers. approach enabled the learn complex features more effectively. For comparison, enhanced pyramid (FT-EPN) benchmarked against Vgg16, Vgg19, Resnet50, InceptionV3, Densenet121, Xception, which were trained transfer (TL) techniques. It also compared next-generation vision transformer (ViT) MaxViT-v2. Results: The achieved an rate 96.77%, outperforming pre-trained TL like ViT Among models, Vgg19 highest at 92.74%. 93.55%, while MaxViT-v2 95.16%. Conclusions: presents FT-EPN classification, offering reference solution future research. provides significant advantages terms simplicity has been evaluated effective applications.

Language: Английский

---

Analysis of N-Way K-Shot Malware Detection Using Few-Shot Learning

Lecture notes in networks and systems, Journal Year: 2023, Volume and Issue: unknown, P. 33 - 44

Published: Jan. 1, 2023

Language: Английский

---

CAMELYON 17 Challenge: A Comparison of Traditional Machine Learning (SVM) with the Deep Learning Method

Wireless Communications and Mobile Computing, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 9

Published: Sept. 30, 2022

The pathologist’s diagnosis is crucial in identifying and categorizing pathological cancer sections, as well the physician’s subsequent evaluation of patient’s condition therapy. It recognised “gold standard”; however, both objective subjective diagnoses have limits, such tissue corruption resulting from nonstandard collection diseased tissue, fixation or delivery, a lack necessary clinical data. In addition, diagnostic pathology encompasses too much information; thus, it requires time effort to grow trained pathologist. Consequently, computer-assisted has become an essential tool for replacing assisting pathologists with computer technology graphical development. this regard, CAMELYON 17 competition was designed identify best algorithm detecting metastases lymph. Each participant given 899 whole-slide photos development their algorithms. More than 300 people enrolled on competition. primarily focused categorization lymph node metastases. TNM classification system primary system. Participants at mostly use learning techniques deep machine learning. order get better understanding top-selected algorithms, we examine advantages limitations traditional classifying breast

Language: Английский

---

Enhancing the Accuracy of an Image Classification Model Using Cross-Modality Transfer Learning

Electronics, Journal Year: 2023, Volume and Issue: 12(15), P. 3316 - 3316

Published: Aug. 2, 2023

Applying deep learning (DL) algorithms for image classification tasks becomes more challenging with insufficient training data. Transfer (TL) has been proposed to address these problems. In theory, TL requires only a small amount of knowledge be transferred the target task, but traditional transfer often presence same or similar features in source and domains. Cross-modality (CMTL) solves this problem by domain completely different from domain, using large data, which helps model learn features. Most existing research on CMTL focused image-to-image transfer. paper, is formulated text domain. Our study started two separately pre-trained models domains obtain network structure. The was via new hybrid (combining BERT BEiT models). Next, GridSearchCV 5-fold cross-validation were used identify most suitable combination hyperparameters (batch size rate) optimizers (SGDM ADAM) our model. To evaluate their impact, 48 two-tuple well-known used. performance evaluation metrics validation accuracy, F1-score, precision, recall. ablation confirms that enhanced accuracy 12.8% compared original addition, results show can significantly impact performance.

Language: Английский

---

Multiclass Classification of Prostate Cancer Gleason Grades Groups Using Features of multi parametric-MRI (mp-MRI) Images by Applying Machine Learning Techniques

Published: Jan. 27, 2023

Prostate cancer (PCa) accounted for 7.8% of all new cases and was the fourth most common in 2020 with 1.4 million cases. With 15.4% newly diagnosed being prostate cancer, it second prevalent type men globally. Due to complex nature PCa, is matter concern that development Computer Aided Diagnosis (CAD) systems detection PCa not keeping up other disciplines. Feature extraction using region interest (ROI) an important step developing CAD systems. Around centre 112 lesions from 99 patients, extracted BVAL, ADC, T2W MRI images. Features based on two three dimensions are ROI. Total 444 features used machine learning classification. Comparison proposed approach feature tested classifiers viz. Support Vector Machine (SVM), Naïve Bayes (NB) k-Nearest Neighbour (KNN). The assessment measures compare aforementioned include accuracy, recall, precision, accuracy as well F1-score, Receiver Operating Characteristics Curve (ROC), AUC, U. Kappa. SVM classification outperform best model ADC modality 44.64 %, FPR 0.1604, PPVGG>1 = 0.7500.

Language: Английский

---

D-UAP: Initially Diversified Universal Adversarial Patch Generation Method

Electronics, Journal Year: 2023, Volume and Issue: 12(14), P. 3080 - 3080

Published: July 14, 2023

With the rapid development of adversarial example technologies, concept patches has been proposed, which can successfully transfer attacks to real world and fool intelligent object detection systems. However, real-world environment is complex changeable, patch attack technology susceptible factors, resulting in a decrease success rate attack. Existing adversarial-patch-generation algorithms have single direction initialization do not fully consider impact initial diversification on its upper limit Therefore, this paper proposes an diversified generation improve effect underlying world. The method uses YOLOv4 as model, experimental results show that adversarial-patch-attack proposed higher than baseline 8.46%, it also stronger fewer training rounds.

Language: Английский

---