<p>Accurate nuclei segmentation from malignant breast cancer histology images plays a crucial role in computer-aided diagnosis and treatment planning. Whereas, using the traditional approaches, complex tissue backdrops and heterogeneity in cell appearance are found to be the major challenges. On this objective, this paper proposed an advanced framework for nuclei segmentation, leveraging an extended convolution block with pyramid pooling-based attention-UNET (ECB-PPA-UNET) model. The proposed model incorporates extended convolution blocks for richer feature extraction and pyramid pooling to capture multi-scale contextual information. An attention mechanism is integrated to focus the model on relevant regions of the image, enhancing the discrimination between nuclei and non-nuclei areas. In experimentation, the ECB-PPA-UNET model demonstrated superior performance over the benchmark models. It achieved a dice coefficient of 0.92, outperforming the standard UNET model by 8%. The precision and recall metrics also showed significant improvements, indicating robustness in segmenting benign nuclei from cancerous ones. As a result, the ECB-PPA-UNET model significantly enhances the accuracy of nuclei segmentation in malignant breast cancer histology images. Its ability to effectively handle variations in nuclear morphology and staining intensity makes it a promising tool for pathological analysis and supports its potential integration into digital pathology workflows for improved diagnostic precision.</p>