Purpose: Accurate identification of tumor margins during surgical resection is essential to minimize the removal of healthy tissue. Hyperspectral imaging (HSI), with its detailed spectral data, enhances the ability to distinguish between cancerous and normal tissues. This study presents a semi-supervised deep learning approach that leverages HSI to support intraoperative decision-making, even with limited annotated data. Materials and Methods: We propose a Semi-Supervised Sparse Autoencoder-Based Distributed Convolutional Neural Network (SAEDCNN) that utilizes both labeled and pseudo labeled HSI data. The model integrates advanced feature extractors such as ResNet-101, local angular patterns (LAP), local phase quantization (LPQ), Ablation-CAM, and VGG-Net within a sparse autoencoder-driven CNN architecture. A novel hybrid attention module, Bottleneck and Efficient Channel Attention (BNEC), is introduced to enhance feature refinement while reducing computational cost. To further improve segmentation accuracy, a hybrid loss function combining Categorical Cross-Entropy (CCE) and hinge loss is employed. Results: SAEDCNN achieved over 97% classification accuracy on benchmark biomedical HSI datasets involving liver bile duct cancer (cholangiocarcinoma) and brain cancer. It performed effectively in segmentation and classification tasks, even with limited supervision. The inclusion of BNEC and the hybrid loss function improved feature learning and model generalization. Conclusion: The combination of HSI and semi-supervised deep learning enables precise tumor detection with reduced reliance on large annotated datasets. SAEDCNN offers high diagnostic accuracy and computational efficiency, making it well-suited for clinical biomedical imaging applications.