Thermal Breast Cancer Detection Using Deep Learning and Grad-CAM Visualization

Salud Ciencia y Tecnología, Journal Year: 2025, Volume and Issue: 5, P. 1518 - 1518

Published: March 18, 2025

This paper presents a robust deep learning framework for thermal breast cancer detection using grayscale images. Leveraging pre-trained VGG16 model, we classify images into 'normal' and 'abnormal' categories, integrating data augmentation techniques to improve model generalization. Grad-CAM visualization elucidates the regions influencing predictions, aiding interpretability. Testing on DMR-IR dataset yielded remarkable AUC-ROC score of 0.97 accuracy exceeding 94%. These findings underscore potential imaging in non-invasive screening, bridging diagnostic with interpretability clinical application.

Language: Английский

---

Evolution of an Artificial Intelligence-Powered Application for Mammography

Diagnostics, Journal Year: 2025, Volume and Issue: 15(7), P. 822 - 822

Published: March 24, 2025

Background: The implementation of radiological artificial intelligence (AI) solutions remains challenging due to limitations in existing testing methodologies. This study assesses the efficacy a comprehensive methodology for performance and monitoring commercial-grade mammographic AI models. Methods: We utilized combination retrospective prospective multicenter approaches evaluate neural network based on Faster R-CNN architecture with ResNet-50 backbone, trained dataset 3641 mammograms. encompassed functional calibration testing, coupled routine technical clinical monitoring. Feedback from testers radiologists was relayed developers, who made updates model. test comprised 112 medical organizations, representing 10 manufacturers mammography equipment encompassing 593,365 studies. evaluation metrics included area under curve (AUC), accuracy, sensitivity, specificity, defects, assessment scores. Results: results demonstrated significant enhancement model's through collaborative efforts among testers, radiologists. Notable improvements functionality, diagnostic stability. Specifically, AUC rose by 24.7% (from 0.73 0.91), accuracy improved 15.6% 0.77 0.89), sensitivity grew 37.1% 0.62 0.85), specificity increased 10.7% 0.84 0.93). average proportion defects declined 9.0% 1.0%, while score 63.4 72.0. Following 2 years 9 months solution integrated into compulsory health insurance system. Conclusions: multi-stage, lifecycle-based substantial potential software integration practice. Key elements this include robust requirements, continuous updates, systematic feedback collection radiologists,

Language: Английский

---

Roles of artificial intelligence and high frame-rate contrast-enhanced ultrasound in the differential diagnosis of Breast Imaging Reporting and Data System 4 breast nodules

Gland Surgery, Journal Year: 2025, Volume and Issue: 14(3), P. 462 - 478

Published: March 1, 2025

Breast cancer prevalence and mortality are rising, emphasizing the need for early, accurate diagnosis. Contrast-enhanced ultrasound (CEUS) artificial intelligence (AI) show promise in distinguishing benign from malignant breast nodules. We compared diagnostic values of AI, high frame-rate CEUS (HiFR-CEUS), their combination Imaging Reporting Data System (BI-RADS) 4 nodules, using pathology as gold standard. Patients with BI-RADS nodules who were hospitalized at Department Thyroid Surgery, Taizhou People's Hospital December 2021 to June 2022 enrolled study.80 female patients (80 lesions) underwent preoperative AI and/or HiFR-CEUS. assessed outcomes HiFR-CEUS, combination, calculating sensitivity (SE), specificity (SP), accuracy (ACC), positive/negative predictive (PPV/NPV). Reliability was Kappa statistics, AI-HiFR-CEUS correlation analyzed Pearson's test. Receiver operating characteristic curves plotted compare combined approach differentiating lesions. Of 80 lesions, 18 pathologically confirmed be benign, while remaining 62 malignant. The SE, SP, ACC, PPV, NPV 75.81%, 94.44%, 80.00%, 97.92%, 53.13% group, 74.20%, 78.75%, 97.91%, 51.51% HiFR-CEUS 98.39%, 88.89%, 96.25%, 96.83%, 94.12% respectively. Thus, group significantly higher than those groups, SP lower (all P<0.05); however, no significant difference found between groups terms PPV (P>0.05). No statistically observed performance P>0.05). had moderate agreement "gold standard" (Kappa =0.551, =0.530, respectively), =0.890). positively correlated (r=0.249, P<0.05). area under (AUCs) both 0.851±0.039, 0.815±0.047, 0.936±0.039, AUC (Z1=2.207, Z2=2.477, respectively, a but not (Z3=0.554, Compared alone or alone, use these two methods our method could further improve guide clinical decision making.

Language: Английский

---

Comparative Evaluation of Machine Learning-Based Radiomics and Deep Learning for Breast Lesion Classification in Mammography

Diagnostics, Journal Year: 2025, Volume and Issue: 15(8), P. 953 - 953

Published: April 9, 2025

Background: Breast cancer is the second leading cause of cancer-related mortality among women, accounting for 12% cases. Early diagnosis, based on identification radiological features, such as masses and microcalcifications in mammograms, crucial reducing rates. However, manual interpretation by radiologists complex subject to variability, emphasizing need automated diagnostic tools enhance accuracy efficiency. This study compares a radiomics workflow machine learning (ML) with deep (DL) approach classifying breast lesions benign or malignant. Methods: matRadiomics was used extract features from mammographic images 1219 patients CBIS-DDSM public database, including 581 cases 638 masses. Among ML models, linear discriminant analysis (LDA) demonstrated best performance both lesion types. External validation conducted private dataset 222 evaluate generalizability an independent cohort. Additionally, EfficientNetB6 model employed comparison. Results: The LDA achieved mean AUC 68.28% 61.53% In external validation, values 66.9% 61.5% were obtained, respectively. contrast, superior performance, achieving 81.52% 76.24% masses, highlighting potential DL improved accuracy. Conclusions: underscores limitations ML-based diagnosis. Deep proves be more effective approach, offering enhanced supporting clinicians improving patient management.

Language: Английский

---

RSDCNet: An efficient and lightweight deep learning model for benign and malignant pathology detection in breast cancer

Digital Health, Journal Year: 2025, Volume and Issue: 11

Published: April 1, 2025

Background Breast cancer is a leading malignant tumor among women globally, with its pathological classification into benign or directly influencing treatment strategies and prognosis. Traditional diagnostic methods, reliant on manual interpretation, are not only time-intensive subjective but also susceptible to variability based the pathologist's expertise workload. Consequently, development of an efficient, automated, precise detection method crucial. Methods This study introduces RSDCNet, enhanced lightweight neural network architecture designed for automatic breast pathology. Utilizing BreakHis dataset, which comprises 9109 microscopic images tumors including various differentiation levels samples, RSDCNet integrates depthwise separable convolution SCSE modules. integration aims reduce model parameters while enhancing key feature extraction capabilities, thereby achieving both design high efficiency. Results demonstrated superior performance across multiple evaluation metrics in task. The achieved accuracy 0.9903, recall 0.9897, F1 score 0.9888, precision 0.9879, outperforming established deep learning models such as EfficientNet, RegNet, HRNet, ViT. Notably, RSDCNet's parameter count stood at just 1,199,662, significantly lower than HRNet's 19,254,102 ViT's 85,800,194, highlighting resource Conclusion presented this excels efficient accurate Compared traditional methods other models, reduces computational consumption offers improved clinical interpretability. advancement provides substantial technical support intelligent diagnosis cancer, paving way more effective planning prognosis assessment.

Language: Английский

---

A Hybrid Model for the Segmentation of Mammogram Images using Otsu Thresholding, Morphology and U-Net

Biomedical & Pharmacology Journal, Journal Year: 2025, Volume and Issue: 18(1), P. 799 - 812

Published: March 31, 2025

Mammogram image segmentation is crucial for early detection and treatment of breast cancer. Timely can help in saving the patient’s life. By accurately identifying isolating regions interest mammograms, we improve diagnostic accuracy. In this paper a hybrid model using Ostu thresholding with morphological operations U-Net proposed accurate mammogram images. The incorporation attention mechanisms residual connections helps enhancing model’s performance. performs better than recent existing models, achieving high precision, recall, F1 score, accuracy, area under curve (AUC). evaluated on MIAS dataset achieved an score 0.9764, precision 0.9802, recall 0.9980, accuracy 0.9902, AUC 0.99997. These results had shown significant improvements comparison making it suitable diagnosis

Language: Английский

---

Benchmarking Deep Learning Algorithms for Breast Cancer Detection: A Comprehensive Review and Evaluation Across Public Imaging Datasets

InfoScience Trends, Journal Year: 2025, Volume and Issue: 2(4), P. 11 - 24

Published: April 14, 2025

Language: Английский

---

AI in 2D Mammography: Improving Breast Cancer Screening Accuracy

Medicina, Journal Year: 2025, Volume and Issue: 61(5), P. 809 - 809

Published: April 26, 2025

Background and Objectives: Breast cancer is a leading global health challenge, where early detection essential for improving survival outcomes. Two-dimensional (2D) mammography the established standard breast screening; however, its diagnostic accuracy limited by factors such as density inter-reader variability. Recent advances in artificial intelligence (AI) have shown promise enhancing radiological interpretation. This study aimed to assess utility of AI lesion classification 2D mammography. Materials Methods: A retrospective analysis was performed on dataset 578 mammographic images obtained from single radiology center. The consisted 36% pathologic 64% normal cases, partitioned into training (403 images), validation (87 test (88 images) sets. Image preprocessing involved grayscale conversion, contrast-limited adaptive histogram equalization (CLAHE), noise reduction, sharpening. convolutional neural network (CNN) model developed using transfer learning with ResNet50. Model performance evaluated sensitivity, specificity, accuracy, area under receiver operating characteristic (AUC-ROC) curve. Results: achieved an overall 88.5% AUC-ROC 0.93, demonstrating strong discriminative capability between cases. Notably, exhibited high specificity 92.7%, contributing reduction false positives improved screening efficiency. Conclusions: AI-assisted holds potential enhance reducing false-positive findings. Although further optimization required minimize negatives. Future efforts should aim improve incorporate multimodal imaging techniques, validate results across larger, multicenter prospective cohorts ensure effective integration clinical workflows.

Language: Английский

---

MedSegBench: A comprehensive benchmark for medical image segmentation in diverse data modalities

Scientific Data, Journal Year: 2024, Volume and Issue: 11(1)

Published: Nov. 25, 2024

MedSegBench is a comprehensive benchmark designed to evaluate deep learning models for medical image segmentation across wide range of modalities. It covers modalities, including 35 datasets with over 60,000 images from ultrasound, MRI, and X-ray. The addresses challenges in imaging by providing standardized train/validation/test splits, considering variability quality dataset imbalances. supports binary multi-class tasks up 19 classes uses the U-Net architecture various encoder/decoder networks such as ResNets, EfficientNet, DenseNet evaluations. valuable resource developing robust flexible algorithms allows fair comparisons different models, promoting development universal tasks. most study among datasets. source code are publicly available, encouraging further research analysis.

Language: Английский

---

Segmentation, classification and interpretation of breast cancer medical images using human-in-the-loop machine learning

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: 37(5), P. 3023 - 3045

Published: Dec. 10, 2024

Language: Английский

---