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.

[Associations of all-cause mortality with admission blood pressure variability during multiple hospitalizations in acute decompensated heart failure]

Objective: To investigate whether admission blood pressure (BP) variability during multiple hospitalizations is associated with all-cause mortality independent of baseline BP in acute decompensated heart failure (ADHF). Methods: Patients with ADHF admitted to the Department of Cardiology, The First Affiliated Hospital of Sun Yat-Sen University from September 2013 to December 2017 were retrospectively enrolled. The risk of all-cause mortality associated with indices of BP variability, including mean admission BPs, standard deviation of BP and coefficient of variation of BP during multiple hospitalizations was assessed, using Cox regression model. Results: A total of 1 006 ADHF patients (mean aged (69.3±13.5) years; 411 (40.8%) female; 670 (66.6%) with preserved ejection fraction) were enrolled. During a median follow-up of 1.54 years, 47.0% of patients died. In all ADHF patients, after adjusting for confounding factors, for every 1-standard deviation (SD) increase in SD and coefficient of variation (CV) of systolic BP, the risk of all-cause mortality increased by 10% and 11%, respectively (SD: HR, 1.10, 95%CI, 1.01-1.21, P=0.029, CV: HR, 1.11, 95%CI, 1.02-1.21, P=0.017); for every 1-SD increase in the mean of diastolic BP, the risk of all cause mortality decreased by 25% (HR, 0.75; 95%CI, 0.65-0.87; P<0.001). In ADHF patients with preserved ejection fraction, after accounted for potential confounders, higher SD and CV of admitted systolic and diastolic BP were significantly associated with higher risk of all-cause mortality, regardless of whether confounding factors were adjusted (P≤0.049); After adjusting for confounding factors, the risk of all-cause mortality increased by 18% and 19% for every 1-SD increase in SD and CV of systolic BP, while the risk of all-cause mortality increased by 11% and 15% for every 1-SD increase in SD and CV of diastolic BP. In ADHF patients with reduced ejection fraction, after adjusting for confounding factors, the higher the mean admission systolic BP during multiple hospitalizations, the lower the risk of total mortality (HR, 0.68; 95%CI, 0.47-1.00; P=0.049). Conclusions: In patients with ADHF, independent of baseline BP, BP variability during multiple hospitalizations was strong predictor of all-cause mortality.

[Contemporary epidemiology and treatment of hospitalized heart failure patients in real clinical practice in China]

Objective: To observe the etiology, comorbidities, clinical features and treatment patterns of hospitalized patients with heart failure (HF) in China. Methods: Data were collected prospectively on hospitalized patients with HF who were enrolled in China Heart Failure Center Registry Study from 169 participating hospitals from January 2017 to August 2018. In this cross-sectional study, patients were stratified by left ventricular ejection fraction (LVEF) category: heart failure with reduced ejection fraction (HFrEF, LVEF<40%); heart failure with mid-ranged ejection fraction (HFmrEF, 40%≤LVEF<50%) and heart failure with preserved ejection fraction (HFpEF, LVEF≥50%). The clinical data were collected, including demographic information, diagnosis, signs, electrocardiogram, echocardiography, laboratory tests, and treatment. Results: A total of 31 356 hospitalized patients with HF were included, 19 072 (60.8%) were males and the average age was (67.9±13.6) years old. The common causes of HF were hypertension (57.2%), coronary heart disease (54.6%), dilated cardiomyopathy (14.7%), valvular heart disease (9.2%). The common complications were atrial fibrillation/atrial flutter (34.1%), diabetes (29.2%), and anemia (26.7%). 32.8% of patients had a history of hospitalization for HF within the previous 12 months. There were 11 034 (35.2%) patients with HFrEF, 6 825 (21.8%) patients with HFmrEF and 13 497 (43.0%) patients with HFpEF. Compared with patients with HFpEF, patients with HFrEF had a lower systolic blood pressure ((124.7±21.1)mmHg(1 mmHg=0.133 kPa) vs. (134.9±22.9)mmHg), faster heart rate ((85±19) beats/minutes vs. (81±19)beats/minutes), and higher percentage of New York Heart Association (NYHA) class Ⅳ, smoking, alcohol, left bundle branch block, and QRS time≥130 ms, and higher levels of blood uric acid, BNP, and NT-proBNP (all P<0.05). Compared with patients with HFmrEF and HFrEF, patients with HFpEF were older, more women, and higher comorbidity burden including hypertension, atrial fibrillation/atrial flutter, anemia and chronic obstructive pulmonary disease (all P<0.05). HFmrEF took a mid-position between HFrEF and HFpEF in age, gender, heart rate, systolic blood pressure, hypertension, atrial fibrillation/atrial flutter, anemia, and chronic obstructive pulmonary disease (all P<0.05). Patients with HFmrEF had the highest proportion of coronary heart disease, myocardial infarction and percutaneous coronary intervention (all P<0.05). During hospitalization, loop diuretics were used in 90.2% of patients, and intravenous inotropics were used in 20.4% of patients. The use of ACEI/ARB/ARNI, β blockers and aldosterone receptor antagonists at discharge were 71.8%, 79.1% and 83.6% in HFrEF and 69.9%, 75.5% and 72.4% in HFmrEF, respectively. The use of digoxin at discharge was 25.3% (HFrEF 36.7%, HFmrEF 23.1%, HFpEF 17.0%). The rates of cardiac resynchronization therapy and implantable cardioverter defibrillator in HFrEF were 2.7% and 2.1%. Conclusions: Among the hospitalized patients with HF in China, coronary heart disease and hypertension are the mostly prevalent causes. HFpEF accounts for a large proportion of hospitalized patients with HF. HFrEF, HFmrEF and HFpEF have different etiology and clinical features. In real-world, there are still large gaps in the effective application of the guideline recommended therapies to HF patients, especially the non-pharmacological therapy option, which needs to be improved further in China.

[The efficiency of 23 G vitrectomy combined with preoperative subtenon injection of triamcinolone acetonide for treatment of retinal detachment associated with choroidal detachment]

Objective: To evaluate the efficiency of 23 G vitrectomy combined with preoperative subtenon injecfion of triamcinolone acetonide for treatment of rhegmatogenous retinal detachment associated with choroidal detachment. Methods: A retrospective analysis. Forty-eight (16 males and 32 females, aged 57.3±13.9) consecutive patients (48 eyes) who were diagnosed with rhegmatogenous retinal detachment associated with choroidal detachment and received 23 G vitrectomy at the Eye Hospital of Wenzhou Medical University during January 2012 and January 2015 were enrolled. Twenty-three eyes were treated with subtenon injection TA 5 d before the planned 23 G vitrectomy (TA group). Twenty-five eyes were treated with dexamethasone 3 to 5 d before the planned vitrectomy (Dex group). Type-B ultrasonic, intraocular pressure, best corrected visual acuity examinations were conducted for all eyes on admission day, preoperatively and at 1 month, 3 month postoperatively, and during the last visit. The rate of reattachment, change of height of choroidal detachment, intraocular pressure, best corrected visual acuity, and the complication of the eyes between the two groups were compared. All patients were followed up at least half a year after the repair surgery. Results: The intraocular pressure of the TA group was higher than the Dex group[(8.58±3.83)mmHg vs. (6.70±2.49)mmHg (1 mmHg=0.133 kPa), (t=2.032)], and the height of choroidal detachment was lower in TA group [0.90(0.00, 3.84)mm vs. 4.03(1.05, 5.38)mm, Z=-2.569, P<0.05]. There is no statistic difference between the reattachment rate of the two groups [95.7%(22/23) vs. 76.0%(19/25), χ(2)=2.304, P=0.129], but it seems it was better in TA group. The best corrected visual acuity results of the last visit was better in TA group than Dex group [(0.91±0.54) vs. (1.25±0.62), t=-2.034, P=0.048]. The rate of hypertention was higher in TA group than Dex group at 1 month, 3 month postoperatively(χ(2)=2.304, 5.648, P<0.05), while there was no statistic difference of hypertention rate during last visit between the two groups (χ(2)=0.006, P=0.941). Conclusions: The treatment of 23 G vitrectomy combined with subtenon injection of triamcinolone acetonide can improve the intraocular pressure, reduce the height of choroidal detachment, and improve the best corrected visual acuity after the surgery, but it may cause heyertenion. (Chin J Ophthalmol, 2018, 54: 252-257).