Abstract-Image denoising is an important pre-processing step in medical image analysis. Different algorithms have been proposed in past three decades with varying denoising performances. More recently, having outperformed all conventional methods, deep learning based models have shown a great promise. These methods are however limited for requirement of large training sample size and high computational costs. In this paper we show that using small sample size, denoising autoencoders constructed using convolutional layers can be used for efficient denoising of medical images. Heterogeneous images can be combined to boost sample size for increased denoising performance. Simplest of networks can reconstruct images with corruption levels so high that noise and signal are not differentiable to human eye.

No abstract

IMPORTANCE There is limited information about the relative effectiveness of cervical cancer screening with primary human papillomavirus (HPV) testing alone compared with cytology in North American populations. OBJECTIVE To evaluate histologically confirmed cumulative incident cervical intraepithelial neoplasia (CIN) grade 3 or worse (CIN3+) detected up to and including 48 months by primary HPV testing alone (intervention) or liquid-based cytology (control). DESIGN, SETTING, AND PARTICIPANTS Randomized clinical trial conducted in an organized Cervical Cancer Screening Program in Canada. Participants were recruited through 224 collaborating clinicians from January 2008 to May 2012, with follow-up through December 2016. Women aged 25 to 65 years with no history of CIN2+ in the past 5 years, no history of invasive cervical cancer, or no history of hysterectomy; who have not received a Papanicolaou test within the past 12 months; and who were not receiving immunosuppressive therapy were eligible. INTERVENTIONS A total of 19 009 women were randomized to the intervention (n = 9552) and control (n = 9457) groups. Women in the intervention group received HPV testing; those whose results were negative returned at 48 months. Women in the control group received liquid-based cytology (LBC) testing; those whose results were negative returned at 24 months for LBC. Women in the control group who were negative at 24 months returned at 48 months. At 48-month exit, both groups received HPV and LBC co-testing. MAIN OUTCOMES AND MEASURES The primary outcome was the cumulative incidence of CIN3+ 48 months following randomization. The cumulative incidence of CIN2+ was a secondary outcome. RESULTS Among 19 009 women who were randomized (mean age, 45 years [10th-90th percentile, 30-59]), 16 374 (8296 [86.9%] in the intervention group and 8078 [85.4%] in the control group) completed the study. At 48 months, significantly fewer CIN3+ and CIN2+ were detected in the intervention vs control group. All Participants Baseline Negative Screen Incidence Rate/1000 (95% CI) at 48 mo Risk Ratio (95% CI) Incidence Rate/1000 (95% CI) at 48 mo Risk Ratio (95% CI) Intervention Group Control Group CIN3+ 2.

Missing data is a significant problem impacting all domains. State-of-the-art framework for minimizing missing data bias is multiple imputation, for which the choice of an imputation model remains nontrivial. We propose a multiple imputation model based on overcomplete deep denoising autoencoders. Our proposed model is capable of handling different data types, missingness patterns, missingness proportions and distributions. Evaluation on several real life datasets show our proposed model significantly outperforms current state-of-the-art methods under varying conditions while simultaneously improving end of the line analytics.

These data provide information that may guide discussions about prognosis between physicians and patients with MBC. In addition, it highlights the importance of stratifying for initial stage at diagnosis in future MBC therapeutic trials.