Scholarlink

<p>Colorectal Cancer (CRC) is a significant contributor to cancer-related mortality worldwide, highlighting the urgent need for effective diagnostic tools. This study addresses this critical challenge by proposing an Explainable Artificial Intelligence (XAI) and Deep Learning (DL)-based framework that integrates multiple transfer learning models, including MobileNetV2, DenseNet-121, ResNet-50, InceptionV3, and EfficientNet-B0, each augmented with customized classification heads. To enhance interpretability, we incorporate XAI techniques, allowing for visual insights into the decision-making processes of the models. Furthermore, we train classical Machine Learning (ML) classifiers on the deep features extracted from Convolutional Neural Networks (CNNs) and apply dimensionality reduction techniques, including Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE), to visualize class separability. Our proposed framework is evaluated using the CRC-VAL-HE-7K histopathological image dataset, with SVM performing promising CNN extracted features. To further establish generalizability, we conducted cross-dataset validation on the NCT-CRC-HE-100K cohort, where SVM again outperformed other models. Moreover, we also present analyses on imbalanced data, ablation analysis, and computational cost. In this scenario, it is noted that SVM is an optimal choice for achieving balanced performance, considering the trade-off between latency and accuracy. This approach not only attains high diagnostic accuracy but also ensures transparency and explainability in results, thereby empowering pathologists to make timely and informed clinical decisions that could significantly improve patient outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep feature-driven SVM model with XAI for reliable colorectal cancer imaging analysis

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Ahmad Almadhor,
  • Stephen Ojo,
  • Thomas I. Nathaniel,
  • Sultan Ahmad,
  • Abdullah Al Hejaili

摘要

Colorectal Cancer (CRC) is a significant contributor to cancer-related mortality worldwide, highlighting the urgent need for effective diagnostic tools. This study addresses this critical challenge by proposing an Explainable Artificial Intelligence (XAI) and Deep Learning (DL)-based framework that integrates multiple transfer learning models, including MobileNetV2, DenseNet-121, ResNet-50, InceptionV3, and EfficientNet-B0, each augmented with customized classification heads. To enhance interpretability, we incorporate XAI techniques, allowing for visual insights into the decision-making processes of the models. Furthermore, we train classical Machine Learning (ML) classifiers on the deep features extracted from Convolutional Neural Networks (CNNs) and apply dimensionality reduction techniques, including Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE), to visualize class separability. Our proposed framework is evaluated using the CRC-VAL-HE-7K histopathological image dataset, with SVM performing promising CNN extracted features. To further establish generalizability, we conducted cross-dataset validation on the NCT-CRC-HE-100K cohort, where SVM again outperformed other models. Moreover, we also present analyses on imbalanced data, ablation analysis, and computational cost. In this scenario, it is noted that SVM is an optimal choice for achieving balanced performance, considering the trade-off between latency and accuracy. This approach not only attains high diagnostic accuracy but also ensures transparency and explainability in results, thereby empowering pathologists to make timely and informed clinical decisions that could significantly improve patient outcomes.