A Novel Dilated Convolution Light Weight Neural Network (DCLW-NN) for the Classification of Breast Thermograms
Breast cancer poses lots of challenges in the medical fraternity worldwide. Hence, it requires early diagnosis to treat the affected people. In this paper, we have presented a novel Dilated Convolution Light Weight Neural Network (DCLW-NN) to analyze the breast thermograms. Convolutional Neural Netw...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11008627/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Breast cancer poses lots of challenges in the medical fraternity worldwide. Hence, it requires early diagnosis to treat the affected people. In this paper, we have presented a novel Dilated Convolution Light Weight Neural Network (DCLW-NN) to analyze the breast thermograms. Convolutional Neural Network (CNN) is a deep learning model to process the hierarchical feature of visual data to conduct image classification. DCLW-NN is a light weight architecture that uses dilation principle in standard convolution to reduce the model parameters and maintain accuracy. To reduce system complexity, dilated Convolution enables the integration of information from a wider range without the need for a large kernel. The proposed model is experimented with two different infrared thermography imaging datasets like DMR-IR and Proprietary Datasets. The model is tested under four cases by training and testing with similar and dissimilar datasets and obtain the classification accuracy. To solve the problem of data scarcity, the model is worked out with datasets at various augmentation level and their performance is compared. Lastly, the categorization layer of the CNN model is replaced with Support Vector Machine(SVM) classifier with L2 regularization and found the prospects of DCLW-NN Classifier. The significant improvement of the model is obtained by varying the hyper tuning parameters of the model. The classifier performance of the proposed model was compared with existing architecture like RESNET-50, AlexNet, VGG-16 and it was found that it produces better accuracy of 99%, Sensitivity of 100% and Specificity of 98%. |
|---|---|
| ISSN: | 2169-3536 |