A Deep Learning Based Methodological Analysis for Breast Cancer Classification

Question

1. What are the challenges in breast cancer diagnosis and prognosis?

2. What DL methods are used for BC detection in ultrasound images?

3. What DL method outperforms others in BC detection using MRI images?

4. How does layered route evolution approach aid BC detection in tomosynthesis images?

Accepted Answer

Breast cancer diagnosis and prognosis present several challenges. One of the main challenges is the detection of early-stage breast cancer, as it can be completely cured if detected in time. Accessing appropriate diagnostic techniques is critical for recognizing the first signs of the disease. Mammography, although commonly used, has limitations in detecting stable breasts. Ultrasound procedures are often employed, but micro masses can be skipped through radiation. Thermography may be more successful than ultrasonography in identifying tiny malignant tumors. Additionally, image processing devices have been developed to aid in the prediction and distinction of malignant and benign lesions. However, inherent challenges such as low contrast, various noises, and a deficiency of awareness by the eye accompanying an image, make the diagnosis and prognosis of breast cancer a complex undertaking. AI, ML, and DL technologies have been utilized to enhance diagnostic efficiency and accuracy, but further research is needed to address these challenges and improve breast cancer detection and prognosis.

Accepted Answer

Several Deep Learning (DL) methods are employed for breast cancer (BC) detection in ultrasound images. These methods include edge extractor strategies like the modified convolutional RNN (CRNN) model proposed by Kim et al., 3D CNNs suggested by Wang et al., and DNNs such as AlexNet and ResNet used by Misra et al. Additionally, Patch-based LeNet, U-Net, and TL strategy with a pre-trained FCN-AlexNet are offered by Yap et al., while DCNN with multi-scale kernels and skip correlations are recommended by Qi et al. Other notable methods include CNN-based techniques by Masud et al., attention-augmented multi-instance (MI) network by Pi et al., and deep CNN approaches like AlexNet and DenseNet201 by Saba et al. Furthermore, Zhang et al. evaluated BC's diagnostic effectiveness using CNN, and Wang et al. proposed CNN with a modified Inception-v3 structure. These methods have shown promising results in terms of accuracy, sensitivity, and specificity in detecting BC in ultrasound images.

Accepted Answer

The KFLI (Knowledgedriven Feature Learning and Integration) framework, a deep network ensemble with domain knowledge, proposed by Feng et al., outperforms other current algorithms in BC detection using MRI images. It achieved 84.6 percent sensitivity, 85.7 percent specificity, and 85.0 percent accuracy in 100 MRI investigations. The SRVFL-AE (The stacked random vector functional link-based autoencoder) by Nayak et al. also showed promising results, with 96.67 percent accuracy and 95.00 percent on the MD-1 and 2 data sets. Additionally, Dalms et al. developed an integrated automated CADe system using DL, and Rasti et al. created ME-CNN (mixed ensemble of CNN) that achieved a precision of 96.39 percent, a sensitivity of 97.73 percent, and a specificity of 94.87 percent. Zhang et al. used CNN and RCNN, with the CLSTM-based RCNN achieving higher accuracy than a regular CNN. Anderson et al. developed two deep TL techniques based on extracting features and fine-tuning them, achieving a 64 percent training, 16 percent validation, and 20 percent testing data set split. Fujioka et al. evaluated the likelihood of malignancy using DenseNet121, InceptionV3, DenseNet169, InceptionResNetV2, InceptionV3, and NasNetMobile, with the top CNN achieving a specificity of 96.0 percent and an AUC of 0.895 for detecting breast MRI.

Accepted Answer

The layered route evolution approach, as created by Samala et al. [58], aims to compress a Deep Convolutional Neural Network (DCNN) for breast cancer (BC) classification while maintaining recognition accuracy. By reducing the number of customizable factors, this approach optimizes the DCNN's performance. In the second step of Tomographic Layering (TL), the DCNN extracts features using feature selection, and classification is performed using Random Forest (RF). This methodology has shown promising results in improving BC detection in tomosynthesis images, making it a valuable technique for researchers and practitioners in the field.

Accepted Answer

DCNN, or Deep Convolutional Neural Network, offers several advantages in breast cancer detection. It provides a large contextual window and significant modeling power, allowing for representation learning to learn features instead of explicitly constructed features used in standard machine learning. This approach can decrease dimensionality without prior data knowledge. However, DCNN also has some disadvantages. It takes a long time to run, which can be a limitation in time-sensitive applications. Additionally, if the testing data does not reflect the training data, DCNN may confound the learning process rather than accurately reflect the dataset's qualities. These factors should be considered when implementing DCNN in breast cancer detection.

Accepted Answer

DL classifies medical images by using deep learning algorithms, such as CNNs, to analyze and extract features from the images. These algorithms require large datasets to achieve high accuracy in classification. The literature review confirms that DL methods, like CNNs, produce higher Reached Accuracy compared to alternative neural network methods. However, the availability of comprehensive datasets is crucial for the efficiency of DL in classifying breast cancer (BC). The review also highlights the importance of image preprocessing, such as denoising and ROI segmentation, to improve accuracy. Additionally, it emphasizes the need for training data equality to avoid bias in the system's results. Overall, DL algorithms have shown promising results in BC diagnosis, but challenges like limited datasets and data augmentation strategies need to be addressed for further advancements.

Accepted Answer

DL requires a massive set of data for training to achieve high precision. However, the vast dataset often has limited data, which poses challenges for training and research. To overcome this, researchers have utilized accessible data from sources like the Cancer Imaging Archive. This data is used to train and evaluate quantitative imaging techniques in the early stages of breast cancer (BC) therapy. By leveraging approaches such as HA-BiRNN, CNN, DLA-EABA, and CRNN, researchers can test and compare the performance of these techniques using various criteria like Accuracy, Sensitivity, Specificity, Precision, and F-Score. The numerical results are presented in tables, and the formulas for calculating these performance metrics are provided. This approach allows for the development and evaluation of advanced imaging techniques for BC diagnosis and treatment.

Accepted Answer

The CRNN enhances Reached Accuracy while reducing computation time and memory usage due to its lightweight network structure. It achieves a 99.75 percent Reached Accuracy, surpassing other algorithms that require many derived components during reduction. This method outperforms current strategies for confirming BC detection findings. Additionally, the CRNN has a sensitivity of 97.21 percent and a specificity of 96.76 percent, making it effective for high-dimensional datasets and BC lesion prediction testing. It also offers excellent sensitivity and specificity for detecting BC mass, potentially improving patient survival rates.

Accepted Answer

The DL algorithm significantly impacts the diagnosis process by lowering the amount of human error, which can lead to erroneous diagnoses. Compared to manual diagnosis, the invention of this algorithm improves diagnostic accuracy in the medical profession. Various DL algorithms have been created to achieve specific goals in BC diagnoses using diverse modalities. The evolution of DL algorithms in medical diagnostics, particularly in BC, was documented in this review study. In the diagnosing process, they all showed a high level of accuracy. This demonstrated that erroneous diagnostics might be avoided by using the DL method. However, the researcher must consider various DL system restrictions; otherwise, the diagnosis will be inaccurate. Further novel DL algorithms may be employed in the future to identify the stage of the BC. Several tuning parameters will be required to successfully train the model for this task, such as dropout rate, kernel functions, and learning rate. These parameters have relied on human specialists in previous years. Finally, an efficient swarm technique is needed to determine the best hyperparameters for the DL algorithm.

A Deep Learning Based Methodological Analysis for Breast Cancer Classification

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What are the challenges in breast cancer diagnosis and prognosis?

2. What DL methods are used for BC detection in ultrasound images?

3. What DL method outperforms others in BC detection using MRI images?

4. How does layered route evolution approach aid BC detection in tomosynthesis images?

5. What are the advantages and disadvantages of DCNN in breast cancer detection?

6. How does DL classify medical images?

7. How does DL require data for training?

8. How does CRNN compare in Reached Accuracy?

9. How does DL algorithm impact BC diagnosis accuracy?

Citations

Scale Invariant Deep Neural Multiple Feature Learning Based Boosted Support Vector Entropy Classification for Breast Cancer Diagnosis using Mammograms

Visual Place Recognition Model using Deep Learning with Arithmetic Optimization Algorithm

Enhanced Breast Cancer Detection From Thermal Images Using <scp>DNN</scp> and Explainable <scp>AI</scp>

References

Machine Learning: Algorithms, Real-World Applications and Research Directions

Automated Breast Ultrasound Lesions Detection Using Convolutional Neural Networks

A Review of Deep Learning in Medical Imaging: Imaging Traits, Technology Trends, Case Studies With Progress Highlights, and Future Promises

Deep Neural Networks Improve Radiologists’ Performance in Breast Cancer Screening

Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis.

Related Papers (5)

Advancements in molecular breast imaging.

Imaging of the Breast

A Tour of Unsupervised Deep Learning for Medical Image Analysis

Exploring the Best Parameters of Deep Learning for Breast Cancer Classification System

Machine Learning Techniques for Biomedical Image Segmentation: An Overview of Technical Aspects and Introduction to State-of-Art Applications