Gastrointestinal tract classification using improved LSTM based CNN
Multimedia Tools and Applications, cilt.79, sa.39-40, ss.28825-28840, 2020 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 79 Sayı: 39-40
- Basım Tarihi: 2020
- Doi Numarası: 10.1007/s11042-020-09468-3
- Dergi Adı: Multimedia Tools and Applications
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, FRANCIS, ABI/INFORM, Applied Science & Technology Source, Compendex, Computer & Applied Sciences, INSPEC, zbMATH
- Sayfa Sayıları: ss.28825-28840
- Anahtar Kelimeler: Colorectal cancer, Deep learning, Endoscopic images, LSTM, Transfer learning
- Ankara Hacı Bayram Veli Üniversitesi Adresli: Evet
Özet
Automated medical image analysis is a challenging field of research that has become quite widespread recently. This process, which is advantageous in terms of both cost and time, is problematic in terms of obtaining annotated data and lack of uniformity. Artificial intelligence is beneficial in the automatic detection of many diseases where early diagnosis is vital for human life. In this study, an effective classification method is presented for a gastrointestinal tract classification task that contains a small number of labeled data and has a sample number of imbalance between classes. According to our approach, using an effective classifier at the end of the convolutional neural network (CNN) structure produces the desired performance even if the CNN structure is not strongly trained. For this purpose, a highly efficient Long Short-Term Memory (LSTM) structure is designed and added to the output of the CNN. Experiments are conducted using AlexNet, GoogLeNet, and ResNet architectures to test the contribution of the proposed approach to the classification performance. Besides, three different experiments are carried out for each architecture where the sample numbers are kept constant as 2500, 5000, and 7500. All experiments are repeated with CNN + ANN and CNN + SVM architectures to compare the performance of our framework. The proposed method has a more successful classification performance than other state-of-the-art methods with 97.90% accuracy.