A segmentation model to detect cevical lesions based on machine learning of colposcopic images

Background

Semantic segmentation is crucial in medical image diagnosis. Traditional deep convolutional neural networks excel in image classification and object detection but fall short in segmentation tasks. Enhancing the accuracy and efficiency of detecting high-level cervical lesions and invasive cancer poses a primary challenge in segmentation model development.

Methods

Between 2018 and 2022, we retrospectively studied a total of 777 patients, comprising 339 patients with high-level cervical lesions and 313 patients with microinvasive or invasive cervical cancer. Overall, 1554 colposcopic images were put into the DeepLabv3+ model for learning. Accuracy, Precision, Specificity, and mIoU were employed to evaluate the performance of the model in the prediction of cervical high-level lesions and cancer.

Results

Experiments showed that our segmentation model had better diagnosis efficiency than colposcopic experts and other artificial intelligence models, and reached Accuracy of 93.29 %, Precision of 87.2 %, Specificity of 90.1 %, and mIoU of 80.27 %, respectively.

Conclution

The DeepLabv3+ model had good performance in the segmentation of cervical lesions in colposcopic post-acetic-acid images and can better assist colposcopists in improving the diagnosis.

[A long-term follow-up study of cardiac resynchronization therapy for children with right ventricle-paced heart failure]

Objective: To evaluate the long-term effects of cardiac resynchronization therapy (CRT) in children with right ventricle-paced heart failure. Methods: Five children with chronically right ventricular-paced heart failure underwent operation of upgrading to CRT in Guangdong Cardiovascular Institute between July 2009 to January 2015. The first time the patients were implanted with endocardial permanent pacemaker was (11.6±4.6) years old. The New York Heart Association (NYHA) functional classification, QRS duration, left ventricular end diastolic diameter (LVDd), left ventricular ejection fraction (LVEF), standard deviation of peak systolic time of left ventricular wall (TS-SD) and follow-up data were retrospectively analyzed. Comparison between pre-and post-operation was performed using paired t test. Results: CRT significantly improved the NYHA class to Ⅰ-Ⅱdegree, reduced the QRS duration ((126±9)vs. (182±21)ms, t=-7.480, P=0.002) and the Z-score of LVDd (2.8±1.1 vs. 4.7±0.9, t=-2.880, P=0.045), and increased the LVEF (43%±10% vs. 28%±6%, t=3.350, P=0.029). No significant difference was found regarding the TS-SD ((48±17)vs. (95±41)ms, t=-2.240, P=0.090) pre- and post-CRT. The longest follow-up period was 9 years. During follow-up, 1 case died of ventricular fibrillation 2 years after upgrading, and 2 cases underwent CRT replacement due to battery depletion 7.2 years and 5.8 years after upgrading, respectively. Conclusion: CRT could be considered for children with chronically right ventricular-paced heart failure and improve heart function significantly.

[Etiological diagnoses in 53 cases of pneumonia]

Secretion was taken from lower respiratory tract for bacterial culture using Japanese made single-sheath catheter brush (SSC) via fiberoptic bronchoscopy in 53 cases with community acquired pneumonia and 16 control cases without pulmonary infection. The results showed that bacteria were isolated in 42 out of the 53 patients, the organisms being pathogenic in 39 out of the 53 (73.5%). Among the bacteria isolated from the 42 cases, gram-negative bacilli accounted for the highest rate of 36% and pneumococcus the next of 31%. There were only 3 cases yielding contamination. It is shown that SSC technique has less chance of contamination and is convenient and practical.