Deep learning-based hysteroscopic intelligent examination and ultrasound examination for diagnosis of endometrial carcinoma

The study aimed to explore the performance of deep learning-based hysteroscopy intelligent examination combined with ultrasound examination in the diagnosis of endometrial carcinoma (EC). Specifically, 80 EC patients, diagnosed by hysteroscopic cervical tissue biopsy were selected as the research subjects, and they were divided into the experimental group, and the control group. The Dense-Pyramid-Attention U-Net (DPA-UNet) algorithm image processing method based on deep learning was applied to diagnose patients in the experimental group. Then, different diagnosis methods were compared for the accuracy rates of the preoperative staging and the diagnosis results. The results showed that compared with U-Net and Dense-Net models, the image clarity processed by the DPA-UNet model was improved and the lesion site was clearer, and its Dice similarity coefficient (DSC), precision, and recall were 80.4 ± 18%, 80.1 ± 15%, 87.6 ± 11%, respectively, higher than those of U-Net and Dense-Net model, and the difference wax statistically significant (P < 0.05); the diagnosis result coincidence rate of the experimental group was 91.8%, significantly better than that of the control group 64.1%, and the difference was statistically significant. In conclusion, the deep learning-based hysteroscope intelligent inspection system combined with ultrasound images may provide an efficient way for early diagnosis of EC.