Classification of skin cancer from dermatological image by multimodal method

Опубликована: Янв. 15, 2025

Skin cancer remains a critical health issue, with over 1.2 million new cases diagnosed annually. Early detection is crucial in reducing mortality rates, yet challenges diagnosis persist due to variability dermoscopic image quality. Traditional methods for skin lesion classification are cumbersome and involve significant manual preprocessing. This study introduces an innovative approach using deep learning automate feature extraction enhance diagnostic accuracy. We investigate ensemble of advanced neural networks (VGG, ResNet, GoogleNet, Visiontransformer) combined multimodal method that integrates patient metadata features. Our dataset includes 9,013 training images 1,002 testing across seven categories pigmented lesions. The approach, mainly the DenseNet121-Mul model, demonstrated superior performance precision, recall, F1-score, achieving F1-score 0.91. findings highlight potentiality multiple deeplearning models combining diverse data types advance accuracy efficiency computer-aided diagnostics dermatology, paving way systems could match dermatologist-level capabilities.

Язык: Английский

---

DTONet a Lightweight Model for Melanoma Segmentation

Bioengineering, Год журнала: 2024, Номер 11(4), С. 390 - 390

Опубликована: Апрель 18, 2024

With the further development of neural networks, automatic segmentation techniques for melanoma are becoming increasingly mature, especially under conditions abundant hardware resources. This allows accuracy to be improved by increasing complexity and computational capacity model. However, a new problem arises when it comes actual applications, as there may not high-end available, in hospitals among general public, who have limited computing In response this situation, paper proposes lightweight deep learning network that can achieve high with minimal resource consumption. We introduce called DTONet (double-tailed octave network), which was specifically designed purpose. Its parameter count is only 30,859, 1/256th mainstream UNet Despite its reduced complexity, demonstrates superior performance terms accuracy, an IOU improvement over other similar models. To validate generalization capability model, we conducted tests on PH2 dataset, results still outperformed existing Therefore, proposed exhibits excellent ability sufficiently outstanding.

Язык: Английский

---

Exploring the Interplay of Dataset Size and Imbalance on CNN Performance in Healthcare: Using X-rays to Identify COVID-19 Patients

Diagnostics, Год журнала: 2024, Номер 14(16), С. 1727 - 1727

Опубликована: Авг. 8, 2024

Introduction: Convolutional Neural Network (CNN) systems in healthcare are influenced by unbalanced datasets and varying sizes. This article delves into the impact of dataset size, class imbalance, their interplay on CNN systems, focusing size training set versus imbalance—a unique perspective compared to prevailing literature. Furthermore, it addresses scenarios with more than two classification groups, often overlooked but prevalent practical settings. Methods: Initially, a was developed classify lung diseases using X-ray images, distinguishing between healthy individuals COVID-19 patients. Later, model expanded include pneumonia To evaluate performance, numerous experiments were conducted varied data sizes imbalance ratios for both binary ternary classifications, measuring various indices validate model’s efficacy. Results: The study revealed that increasing positively impacts this improvement saturates beyond certain size. A novel finding is balance ratio influences performance significantly behavior three-class mirrored classification, underscoring importance balanced accurate classification. Conclusions: emphasizes fact achieving representation crucial optimal healthcare, challenging conventional focus Balanced improve accuracy, two-class scenarios, highlighting need data-balancing techniques reliability effectiveness. Motivation: Our motivated scenario 100 patient samples, offering options: 200 samples an 500 (400 individuals). We aim provide insights choice based enriching discourse stakeholders interested performance. Limitations: Recognizing single generalizability limitations, we assert further studies diverse needed.

Язык: Английский

---

Multidirectional Analysis of Curvelet Against Skin Cancer

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Янв. 24, 2024

An elevated chance of getting another melanoma is associated with a personal history the disease. Individuals who have already had 2–5% probability one later. Compared to individuals whose initial was superficial spreading melanoma, those first lentigo maligns or nodular are at greater peril emerging secondary dominant cancer. Melanoma risk double in special antiquity squamous cell carcinoma. The likelihood doubled if you particular times past basal In addition, higher persons actinic keratosis than without condition. automated technique for classifying skin cancer, proposed this work. image gathered data used as input system, and various handling methods remain smeared improve picture's characteristics. curvelet separate benign from malignant cancer collect relevant these pictures so that classifier may be trained tested. basic wrapper curvelet's coefficients characteristics utilized classification. Curvelet works well images cartoon edges aligned textures. database digital photos, three-layer back-propagation neural network has 75.6% recognition accuracy.

Язык: Английский

---

Concurrent Learning Approach for Estimation of Pelvic Tilt from Anterior–Posterior Radiograph

Bioengineering, Год журнала: 2024, Номер 11(2), С. 194 - 194

Опубликована: Фев. 17, 2024

Accurate and reliable estimation of the pelvic tilt is one essential pre-planning factors for total hip arthroplasty to prevent common post-operative complications such as implant impingement dislocation. Inspired by latest advances in deep learning-based systems, our focus this paper has been present an innovative accurate method estimating functional (PT) from a standing anterior–posterior (AP) radiography image. We introduce encoder–decoder-style network based on concurrent learning approach called VGG-UNET (VGG embedded U-NET), where fully convolutional known VGG at encoder part image segmentation network, i.e., U-NET. In bottleneck VGG-UNET, addition decoder path, we use another path utilizing light-weight connected layers combine all extracted feature maps final convolution layer thus regress PT. test phase, exclude consider only single target task PT estimation. The absolute errors obtained using VGG, Mask R-CNN are 3.04 ± 2.49, 3.92 2.92, 4.97 3.87, respectively. It observed that leads more prediction with lower standard deviation (STD). Our experimental results demonstrate proposed multi-task significantly improved performance compared best-reported cascaded networks.

Язык: Английский

---

Stereotactic Needle Path Planning for Brain Biopsy Based on Volumetric Segmentation of MRI and Heuristic Approach

Lecture notes in networks and systems, Год журнала: 2024, Номер unknown, С. 473 - 487

Опубликована: Окт. 26, 2024

Язык: Английский

---