Enhanced Breast Tumor Detection with a CNN-LSTM Hybrid Approach: Advancing Accuracy and Precision
Shiva Mehta, Saniya Khurana · 2024
Since most breast cancers can be treated, early detection and accurate disease classification are critical. This paper introduces the architecture of a new model developed by integrating a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) under the federated learning model. The objective is to categorize breast cancers into five distinct categories accurately: They include benign, malignant, cystic, fibroadenoma, and phyllodes tumors. A pervasive and comprehensive set of data was gathered to evaluate five different medical institutes. Relative to the entail standard CNN model, our model achieved a high accuracy level of 93%. 0%, precision of 92. 5%, Recall of 93. Specifically, Accuracy was at 0%, and an F1-score of 92-7%. Meanwhile, the other traditional CNN models obtained an average accuracy of 88. 5%, precision of 87. 9%, Recall of 88. 47%, while the corresponding Recall for the proposed method is 92. 54% in class 0, 66. 67% in classes 1, 7. 50 % in class 2, and F1-score is 81. 0%. This made it possible for the universities to cooperate in developing the systems while maintaining the confidentiality of data, an essential factor in the medical field. An ablation study focused on the importance of each component; specifically, the CNN-LSTM model achieved a 92%-5% accuracy with a plain Abi word list without federated learning and 93 0% with its inclusion. However, the research also understands limitations; for example, the dataset is small, and adding FL requires extra computation.