A Hybrid Convolutional Neural Network and Long Short-Term Memory (Cnn-Lstm)With Attention Architecture for Precise Medical Image Analysis and Disease Diagnosis

Analyzing medical images is an essential aspect of computer-aided diagnosis in the modern era. It facilitates timely detection and accurate identification of diseases. Nevertheless, traditional deep learning approaches encounter difficulties in consistently performing across different imaging modalities because of challenges including: high intra-class variability, image noise, and a lack of labeled data for supervised learning. This work proposes a Hybrid CNN-LSTM-Attention architecture to address these challenges, blending i) the spatio-temporal learning capability of CNNs, ii) the temporal modeling ability of LSTM networks, and iii) the contextual focusing capability of an Attention module. The CNN used in the proposed hybrid architecture learns deep hierarchical spatio-temporal features, the LSTM networks learn sequential dependencies, and the Attention module learn to focus on areas significant for diagnosis, leading to improvements in both accuracy, and interpretability of the model. The system evaluated the hybrid CNN-LSTM-Attention framework across several medical imaging benchmarks, including Chest X-Ray (Pneumonia vs. Normal), MRI Brain Tumor Classification and Retinal OCT. The proposed architecture achieved 95.62 % accuracy, 94.78 % precision, 95.03 % recall, and$94.90 \% ~\mathrm{F} 1$-score, which outperformed baseline CNN and CNN-LSTM baseline models, as well as transformer architectures.