MSCC: Multi-Class Skin Cancer Classification and Interpretable Deep Learning Systems

Опубликована: Май 2, 2024

Nowadays, skin cancer is a common and potentially deadly disease, requiring prompt precise diagnosis for effective treatment. Our study introduces multi-class classification (MSCC) system employing deep convolutional neural networks (DCNNs) interpretable learning frameworks. This approach enhances accuracy speed, providing clear, understandable visual explanations. By addressing challenges like feature extraction from irregular, artifactladen images improving generalization interpretability, this promises to significantly aid in early detection, thus saving lives reducing the strain on healthcare professionals. The effectiveness of proposed model assessed using ISIC-2018 ISIC-2019 datasets imaging. successfully distinguishes between seven types lesions-benign ker-atosis lesions, melanoma, basal cell carcinoma, melanocytic nevi, vascular actinic keratosis, dermatofibroma-with high accuracy, precision, recall, F1 score, all averaging at 96.37%, 96.39%, 96.35%, 96.36%, respectively. To delve deeper into model's predictions, we employ local in-terpretable model-agnostic explanations (LIME) framework SHapley Additive exPlanations (SHAP) values. These techniques generate aligned with prior beliefs adhere best practices general incorporation explainability utility real clinical scenarios.

Язык: Английский

---

A Federated Learning-based Model for the Detection of Lung Cancer from CT Scan Images

Опубликована: Май 2, 2024

The lung is an important organ of the human body. This can be affected by different types diseases. Lung cancer one them, and it most lethal cancers. Early faster detection this disease reduce its spread in In study, a privacy-preserving, federated learning-based approach has been proposed to detect from CT scan images. For that first, dataset collected, which contains four classes: adenocarcinoma, large-cell carcinoma, normal, squamous-cell carcinoma Secondly, various preprocessing techniques have applied Then, third step, Transfer Learning (TL)-based models, are: MobileNet, MobileNetV2, ResNet50V2, VGG16, InceptionV3, implemented find optimal model. Among MobileNetV2 achieved highest accuracy 92.27%. next last Federated (FL)-based model developed using Learning-based outperformed conventional terms performance. It accuracy, precision, recall, f-1 score 93.92 %, 93.50 93.50%, 93.25%, respectively. Nevertheless, approach, not only performance increased, but also users did need share their data. So, method ensure privacy data shared hospitals or clinics.

Язык: Английский

---

Hybrid CNN Model for Pulmonary Disease Detection

Опубликована: Апрель 2, 2024

This research proposes a hybrid convolutional neural network (CNN) model for detecting various pulmonary diseases using substantial dataset of Lung CT-Scan images. The architecture integrates ResNet, DenseNet-121, and InceptionV3 to harness diverse feature extraction capabilities, targeting the identification like AdenoCarcinoma, Large Cell Carcinoma, Squamous COVID-19, Normal cases. goal is enhance accuracy sensitivity in disease early diagnosis intervention.The CNN undergoes training on extensive dataset, utilizing transfer learning techniques leverage pre-trained weights from InceptionV3. process fine-tunes model, enabling it capture intricate patterns indicative present images, with specific focus distinguishing between different categories.Evaluation an independent test demonstrates model's efficacy, exhibiting improved performance compared individual models. achieves average 98.61% loss 0.0971 training, 81.70% 0.4649 validation. In phase, attains 84.40% 0.4387. Preliminary results suggest that provides enhanced detection, particularly classification amalgamation architectures enhances ability recognize both subtle prominent associated conditions. contributes significantly advancing automated diagnostic tools diseases, aiming facilitate detection improve overall healthcare outcomes.

Язык: Английский

---

An Improved Archimedes Optimization-aided Multi-scale Deep Learning Segmentation with dilated ensemble CNN classification for detecting lung cancer using CT images

Network Computation in Neural Systems, Год журнала: 2024, Номер unknown, С. 1 - 39

Опубликована: Июль 8, 2024

Early detection of lung cancer is necessary to prevent deaths caused by cancer. But, the identification in lungs using Computed Tomography (CT) scan based on some deep learning algorithms does not provide accurate results. A novel adaptive developed with heuristic improvement. The proposed framework constitutes three sections as (a) Image acquisition, (b) Segmentation Lung nodule, and (c) Classifying raw CT images are congregated through standard data sources. It then followed nodule segmentation process, which conducted Adaptive Multi-Scale Dilated Trans-Unet3+. For increasing accuracy, parameters this model optimized proposing Modified Transfer Operator-based Archimedes Optimization (MTO-AO). At end, segmented subjected classification procedure, namely, Advanced Ensemble Convolutional Neural Networks (ADECNN), it constructed Inception, ResNet MobileNet, where hyper tuned MTO-AO. From networks, final result estimated high ranking-based classification. Hence, performance investigated multiple measures compared among different approaches. Thus, findings demonstrate prove system's efficiency detecting help patient get appropriate treatment.

Язык: Английский

---

An enhanced Garter Snake Optimization-assisted deep learning model for lung cancer segmentation and classification using CT images

Journal of Medical Engineering & Technology, Год журнала: 2024, Номер 48(4), С. 121 - 150

Опубликована: Май 18, 2024

An early detection of lung tumors is critical for better treatment results, and CT scans can reveal lumps in the lungs which are too small to be picked up by conventional X-rays. imaging has advantages, but it also exposes a person radiation from ions, raises possibility malignancy, particularly when procedure done. Access expensive-quality related sophisticated analytic tools might restricted environments with fewer resources due their high cost limited availability. It will need an array creative technological innovations overcome such weaknesses. This paper aims design heuristic deep learning-aided cancer classification using images. The collected images undergone segmentation, performed Shuffling Atrous Convolutional (SAC) based ResUnet++ (SACRUnet++). Finally, Adaptive Residual Attention Network (ARAN) inputting segmented Here parameters ARAN optimally tuned Improved Garter Snake Optimization Algorithm (IGSOA). developed performance compared models showed accuracy.

Язык: Английский

---