<p>Cervical cancer is the most prevalent malignant tumor of the female reproductive system, which seriously threatens the life and health of patients. The damage to the reproductive system caused by it is an important trigger for infertility in women of childbearing age. Deep learning-based pathology image intelligent analysis technology provides a new way to improve the accuracy of early diagnosis. However, the jump connection mechanism in the traditional residual structure is prone to feature dilution, resulting in the loss of key spatial details in pathological images, which seriously affects the quantitative accuracy of the morphological features of cell nuclei; the expansion of the receptive field caused by the deepening of the network layer will result in the blurring of the boundaries of the tiny lesions, which is particularly unfavorable to the precise localization of early cancerous regions. To address the above challenges, this study proposes a multi-task learning-based method for assisted diagnosis of cervical cancer (MTLCS), whose core innovations include: (1) A multi-task coordinated architecture is constructed, which achieves feature reuse to improve efficiency through joint segmentation-classification optimization and reduces computation through parameter sharing; (2) Constructing a feature reorganization module, which enhances the interaction between features by disrupting and reorganizing them; (3) A dual pooling module is constructed to better extract detail information by fusing maximum pooling and average pooling. Experiments on cervical cancer show that MTLCS achieves a DSC coefficient of 81.3% and a classification accuracy of 97.9% in the nucleus segmentation task. The system provides a reliable and intelligent auxiliary tool for clinical pathology diagnosis, and has important clinical value for improving patient prognosis and protecting reproductive function.</p>