Towards Linking CNN Decisions with Cancer Signs for Breast Lesion Classification from Ultrasound Images

Eskandari, Ali; Du, Hongbo; AlZoubi, Alaa

doi:10.1007/978-3-030-80432-9_32

Cited by 3 publications

(4 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Grad-CAM, a gradient-weighted class activation mapping scheme, was developed to generalize CAM without requiring a GAP layer [8]. EGrad-CAM [21] further uses the entropy of feature maps as a measure to select and only visualize feature maps with a high amount of information. Another approach of utilizing the gradient-free methods to visualize CNN is introduced by Score-CAM [22], Ablation-CAM [23], and Clustered-CAM [24].…”

Section: Explainability Of Dcnn Model Decisionsmentioning

confidence: 99%

“…To gain insight into the classification decisions made by the DCNN models for distinguishing between benign and malignant breast lesions, we employ different visualization techniques: in this study, specifically EGrad-CAM [21] and Ablation-CAM [23]. By examining the generated heatmaps as described in Section 2.3, we gained insights into the importance of specific features or regions in the image that contribute to the overall classification decision made by the DCNN models.…”

Section: Cnn Classification Decision Visualizationmentioning

confidence: 99%

“…EGrad-CAM: EGrad-CAM [21] calculates the Shannon entropy of each feature map in the desired layer and generates a heatmap based only on the feature maps with substantial information that significantly contributes towards the classification decision of BNet and GNet models. We applied Equations ( 2) and ( 3) to generate a heatmap for both the relu5_4' and inception_5b-output' layers of the BNet and GNet models, respectively.…”

Section: Cnn Classification Decision Visualizationmentioning

confidence: 99%

“…Subsequently, we further scrutinize the heatmap scores of various cancer indications in the lesion image's corresponding areas and their contributions to the CNN decision. By leveraging the ground-truth labels of the pixels or regions representing cancer characteristics, we analyze the heatmap scores of the corresponding characteristic locations for different prediction outcomes: True Negative (TN, i.e., correctly classified benign cases), True Positive (TP, i.e., correctly classified malignant cases), False Positive (FP, i.e., misclassified EGrad-CAM: EGrad-CAM [21] calculates the Shannon entropy of each feature map in the desired layer and generates a heatmap based only on the feature maps with substantial information that significantly contributes towards the classification decision of BNet and GNet models. We applied Equations ( 2) and (3) to generate a heatmap for both the 'relu5_4' and 'inception_5b-output' layers of the BNet and GNet models, respectively.…”

Section: Explaining Cnn Classification Decision Using Lesion Characte...mentioning

confidence: 99%

See 3 more Smart Citations

Explainable DCNN Decision Framework for Breast Lesion Classification from Ultrasound Images Based on Cancer Characteristics

AlZoubi,

Eskandari,

et al. 2024

Bioengineering

Self Cite

View full text Add to dashboard Cite

In recent years, deep convolutional neural networks (DCNNs) have shown promising performance in medical image analysis, including breast lesion classification in 2D ultrasound (US) images. Despite the outstanding performance of DCNN solutions, explaining their decisions remains an open investigation. Yet, the explainability of DCNN models has become essential for healthcare systems to accept and trust the models. This paper presents a novel framework for explaining DCNN classification decisions of lesions in ultrasound images using the saliency maps linking the DCNN decisions to known cancer characteristics in the medical domain. The proposed framework consists of three main phases. First, DCNN models for classification in ultrasound images are built. Next, selected methods for visualization are applied to obtain saliency maps on the input images of the DCNN models. In the final phase, the visualization outputs and domain-known cancer characteristics are mapped. The paper then demonstrates the use of the framework for breast lesion classification from ultrasound images. We first follow the transfer learning approach and build two DCNN models. We then analyze the visualization outputs of the trained DCNN models using the EGrad-CAM and Ablation-CAM methods. We map the DCNN model decisions of benign and malignant lesions through the visualization outputs to the characteristics such as echogenicity, calcification, shape, and margin. A retrospective dataset of 1298 US images collected from different hospitals is used to evaluate the effectiveness of the framework. The test results show that these characteristics contribute differently to the benign and malignant lesions’ decisions. Our study provides the foundation for other researchers to explain the DCNN classification decisions of other cancer types.

show abstract

Section: Explainability Of Dcnn Model Decisionsmentioning

confidence: 99%

Section: Cnn Classification Decision Visualizationmentioning

confidence: 99%

Section: Cnn Classification Decision Visualizationmentioning

confidence: 99%

Section: Explaining Cnn Classification Decision Using Lesion Characte...mentioning

confidence: 99%

See 2 more Smart Citations

Explainable DCNN Decision Framework for Breast Lesion Classification from Ultrasound Images Based on Cancer Characteristics

AlZoubi,

Eskandari,

et al. 2024

Bioengineering

Self Cite

View full text Add to dashboard Cite

show abstract

What and Why? Interpretability in Colon Cancer Detection

Pal

Kar

Sekh³

2023

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

Classification of breast lesions in ultrasound images using deep convolutional neural networks: transfer learning versus automatic architecture design

AlZoubi,

Lu,

Zhu

et al. 2023

Med Biol Eng Comput

Self Cite

View full text Add to dashboard Cite

Deep convolutional neural networks (DCNNs) have demonstrated promising performance in classifying breast lesions in 2D ultrasound (US) images. Exiting approaches typically use pre-trained models based on architectures designed for natural images with transfer learning. Fewer attempts have been made to design customized architectures specifically for this purpose. This paper presents a comprehensive evaluation on transfer learning based solutions and automatically designed networks, analyzing the accuracy and robustness of different recognition models in three folds. First, we develop six different DCNN models (BNet, GNet, SqNet, DsNet, RsNet, IncReNet) based on transfer learning. Second, we adapt the Bayesian optimization method to optimize a CNN network (BONet) for classifying breast lesions. A retrospective dataset of 3034 US images collected from various hospitals is then used for evaluation. Extensive tests show that the BONet outperforms other models, exhibiting higher accuracy (83.33%), lower generalization gap (1.85%), shorter training time (66 min), and less model complexity (approximately 0.5 million weight parameters). We also compare the diagnostic performance of all models against that by three experienced radiologists. Finally, we explore the use of saliency maps to explain the classification decisions made by different models. Our investigation shows that saliency maps can assist in comprehending the classification decisions. Graphical Abstract

show abstract

Towards Linking CNN Decisions with Cancer Signs for Breast Lesion Classification from Ultrasound Images

Cited by 3 publications

References 23 publications

Explainable DCNN Decision Framework for Breast Lesion Classification from Ultrasound Images Based on Cancer Characteristics

Explainable DCNN Decision Framework for Breast Lesion Classification from Ultrasound Images Based on Cancer Characteristics

What and Why? Interpretability in Colon Cancer Detection

Classification of breast lesions in ultrasound images using deep convolutional neural networks: transfer learning versus automatic architecture design

Contact Info

Product

Resources

About