A Study of CNN and Transfer Learning in Medical Imaging: Advantages, Challenges, Future Scope

Sustainability, Journal Year: 2023, Volume and Issue: 15(7), P. 5930 - 5930

Published: March 29, 2023

This paper presents a comprehensive study of Convolutional Neural Networks (CNN) and transfer learning in the context medical imaging. Medical imaging plays critical role diagnosis treatment diseases, CNN-based models have demonstrated significant improvements image analysis classification tasks. Transfer learning, which involves reusing pre-trained CNN models, has also shown promise addressing challenges related to small datasets limited computational resources. reviews advantages imaging, including improved accuracy, reduced time resource requirements, ability address class imbalances. It discusses challenges, such as need for large diverse datasets, interpretability deep models. What factors contribute success these networks? How are they fashioned, exactly? motivated them build structures that did? Finally, current future research directions opportunities, development specialized architectures exploration new modalities applications using techniques. Overall, highlights potential field while acknowledging continued overcome existing limitations.

Language: Английский

---

A systematic review of intermediate fusion in multimodal deep learning for biomedical applications

Image and Vision Computing, Journal Year: 2025, Volume and Issue: unknown, P. 105509 - 105509

Published: March 1, 2025

Language: Английский

---

Incorporating a Novel Dual Transfer Learning Approach for Medical Images

Sensors, Journal Year: 2023, Volume and Issue: 23(2), P. 570 - 570

Published: Jan. 4, 2023

Recently, transfer learning approaches appeared to reduce the need for many classified medical images. However, these still contain some limitations due mismatch of domain between source and target domain. Therefore, this study aims propose a novel approach, called Dual Transfer Learning (DTL), based on convergence patterns domains. The proposed approach is applied four pre-trained models (VGG16, Xception, ResNet50, MobileNetV2) using two datasets: ISIC2020 skin cancer images ICIAR2018 breast images, by fine-tuning last layers sufficient number unclassified same disease small task, in addition data augmentation techniques balance classes increase samples. According obtained results, it has been experimentally proven that improved performance all models, where without augmentation, VGG16 model, Xception ResNet50 MobileNetV2 model are 0.28%, 10.96%, 15.73%, 10.4%, respectively, while, with 19.66%, 34.76%, 31.76%, 33.03%, respectively. highest compared rest when classifying dataset, as 96.83%, 96.919%, 96.826%, 96.825%, 99.07%, 94.58% accuracy, precision, recall, F1-score, sensitivity, specificity To classify ICIAR 2018 dataset cancer, 99%, 99.003%, 98.995%, 98.55%, 99.14% specificity, Through models' was performed disease.

Language: Английский

---

Lung Tumor Image Segmentation from Computer Tomography Images Using MobileNetV2 and Transfer Learning

Bioengineering, Journal Year: 2023, Volume and Issue: 10(8), P. 981 - 981

Published: Aug. 20, 2023

Background: Lung cancer is one of the most fatal cancers worldwide, and malignant tumors are characterized by growth abnormal cells in tissues lungs. Usually, symptoms lung do not appear until it already at an advanced stage. The proper segmentation cancerous lesions CT images primary method detection towards achieving a completely automated diagnostic system. Method: In this work, we developed improved hybrid neural network via fusion two architectures, MobileNetV2 UNET, for semantic from images. transfer learning technique was employed pre-trained utilized as encoder conventional UNET model feature extraction. proposed efficient approach that performs lightweight filtering to reduce computation pointwise convolution building more features. Skip connections were established with Relu activation function improving convergence connect layers MobileNetv2 decoder allow concatenation maps different resolutions decoder. Furthermore, trained fine-tuned on training dataset acquired Medical Segmentation Decathlon (MSD) 2018 Challenge. Results: tested evaluated 25% obtained MSD, achieved dice score 0.8793, recall 0.8602 precision 0.93. It pertinent mention our outperforms current available networks, which have several phases testing.

Language: Английский

---

Automated Ultrasonography of Hepatocellular Carcinoma using Discrete Wavelet Transform based Deep-learning Neural Network

Medical Image Analysis, Journal Year: 2025, Volume and Issue: 101, P. 103453 - 103453

Published: Jan. 5, 2025

Language: Английский

---

Enhancing Medical Image Reclamation for Chest Samples Using B-Coefficients, DT-CWT and EPS Algorithm

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 113360 - 113375

Published: Jan. 1, 2023

This paper introduces a novel approach for medical image reclamation, specifically focusing on enhancing chest resolution. The proposed method the Dual-Tree Complex Wavelet Transform (DT-CWT) with Edge Preservation Smoothing (EPS) filters to balance visual clarity. resulting Image Reclamation system maintains high-quality results while preserving edges. Performance validation using established metrics like Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Root Mean Square Error (RMSE), and entropy demonstrates substantial improvements: PSNR of 31, SSIM 0.99, RMSE 8.25, 1.03. Furthermore, algorithm extracts features from enhanced through symlet transform, allowing Bhattacharya coefficient computation unique bin analysis enhance retrieval. Experimental show efficiency gains, increasing top 5 matching images' retrieval score 320 512. promises reclamation in emergency settings, facilitating quicker more accurate diagnoses treatments acute injuries. Ultimately, this work can potentially save lives, reduce complications, improve patient outcomes trauma emergencies.

Language: Английский

---

An Alzheimer’s disease classification model using transfer learning Densenet with embedded healthcare decision support system

Decision Analytics Journal, Journal Year: 2023, Volume and Issue: 9, P. 100348 - 100348

Published: Oct. 29, 2023

Training a Convolutional Neural Network (CNN) from scratch is time-consuming and expensive. In this study, we propose implementing the DenseNet architecture for classification of AD in three classes. Our approach leverages transfer learning architectures as base model showcases superior performance on MRI dataset compared to other techniques. We use variety methodologies provide thorough study our model. first create baseline without data augmentation, addressing difficulties classifying Alzheimer's disease (AD) caused by high-dimensional brain scans. The improved obtained through augmentation then highlighted, demonstrating its effectiveness handling sparse assisting generalization. also investigate impact omitting particular transformations modifying split ratios, providing more insights into behavior Through comprehensive evaluation, demonstrate that proposed system achieves an accuracy 96.5% impressive AUC 99%, surpassing previous methods. This mainly highlights architecture, current limitations future recommendation. Moreover, incorporating healthcare decision support further aid valuable diagnosis decision-making clinical settings.

Language: Английский

---

Lung Diseases Diagnosis-Based Deep Learning Methods: A Review

Journal of Techniques, Journal Year: 2023, Volume and Issue: 5(3), P. 158 - 173

Published: Sept. 25, 2023

This review paper examines the current state of lung disease diagnosis based on deep learning (DL) methods. Lung diseases, such as Pneumonia, TB, Covid-19, and cancer, are significant causes morbidity mortality worldwide. Accurate timely these diseases is essential for effective treatment improved patient outcomes. DL methods, which utilize artificial neural networks to extract features from medical images automatically, have shown great promise in improving accuracy efficiency diagnosis. discusses various methods that been developed diagnosis, including convolutional (CNNs), (DNNs), generative adversarial (GANs). The advantages limitations each method discussed, along with types imaging techniques used, X-ray computed tomography (CT). In addition, most commonly used performance metrics evaluating diagnosis: area under curve (AUC), sensitivity, specificity, F1-score, accuracy, precision, receiver operator characteristic (ROC). Moreover, challenges using limited availability annotated data, variability presentation, interpretability generalizability models, highlighted this paper. Furthermore, strategies overcome challenges, transfer learning, data augmentation, explainable AI, also discussed. concludes a call further research address remaining realize DL's full potential treatment.

Language: Английский

---

A deep ensemble learning method for cherry classification

European Food Research and Technology, Journal Year: 2024, Volume and Issue: 250(5), P. 1513 - 1528

Published: March 11, 2024

Abstract In many agricultural products, information technologies are utilized in classification processes at the desired quality. It is undesirable to mix different types of cherries, especially export-type cherries. this study on one important export products Turkey, cherry species was carried out with ensemble learning methods. study, a new dataset consisting 3570 images seven grown Isparta region created. The generated trained six deep models pre-learning original and incremental dataset. As result training data, best obtained from DenseNet169 model an accuracy 99.57%. two results were transferred 100% rate Maximum Voting model.

Language: Английский

---

Automated classification and localization of sewer pipe defects in small-sample CCTV imagery: an enhanced transformer-based framework

Journal of Civil Structural Health Monitoring, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 15, 2025

Language: Английский

---