Background 
Lactation support in transgender women is increasingly recognized as clinically important yet has limited clinical precedent. While previous case reports describe successful lactation induction their protocols involve domperidone, a galactagogue unavailable for this purpose in the US and recommended against by the US Food and Drug Administration (FDA). We present a case of successful lactation induction in a transgender woman using a protocol not involving galactagogues. 
Case Presentation 
Our patient was a 36-year-old transgender woman, a United States military veteran 
receiving care through the Veterans Health Administration (VHA), who was planning on having a child with her partner and wanted to breastfeed. Her partner, a transgender man, carried and delivered the child and wished to return to testosterone use shortly after delivery. Her primary goals included sufficient milk production to allow bonding, even if not the sole source of infant nutrition. Our multidisciplinary team included a clinical pharmacist with gender-affirming hormone therapy expertise and a lactation specialist with gender-affirming therapy expertise. Our patient was status post orchiectomy with an initial testosterone level of 27 ng/dL, and on a dose of 4mg PO 17-beta estradiol. We began the patient’s lactation induction 98 days prior to her partner’s pregnancy and 377 days prior to her 
partner’s eventual due date. Our protocol involved 1) increased estrogen and 
progesterone to mimic physiologic levels during pregnancy; 2) physical nipple stimulation using a breast pump; and 3) reduced estrogen and progesterone shortly before delivery to mimic physiologic levels after birth. We opted against a galactagogue due to FDA recommendations against domperidone as well as patient concerns of side effects with other galactagogues. Interestingly, our patient’s estradiol levels remained under 200 pg/mL throughout the lactation induction process, even at higher levels of 17-beta estradiol dosing. This is in consonance with other published case-reports on lactation-induction in transwomen. 
Discussion 
Working with a multidisciplinary team including pharmacy and lactation specialists and using a protocol involving (1) initial increase of estradiol and progesterone, (2) nipple stimulation, and (3) eventual reduction of estradiol and progesterone, our patient was able to lactate without the use of galactagogues, allowing her to meet her goal of bonding with her baby. This is the first reported case of successful lactation induction without galactagogues for a transgender woman to our knowledge, and provides an example protocol modeled on physiologic lactation stimuli that other clinicians may adapt with their own patients in similar circumstances.

Inducing lactation in a transgender woman without galactagogues: A case report

Acute osteomyelitis in the setting of chronic steroid therapy for the treatment of 17-alpha-hydroxylase deficient congenital adrenal hyperplasia

Unsupervised image segmentation aims to match low-level visual features with semantic-level representations without outer supervision. In this paper, we address the critical properties from the view of feature alignments and feature uniformity for UISS models. We also make a comparison between UISS and image-wise representation learning. Based on the analysis, we argue that the existing MI-based methods in UISS suffer from representation collapse. By this, we proposed a robust network called Semantic Attention Network(SAN), in which a new module Semantic Attention(SEAT) is proposed to generate pixel-wise and semantic features dynamically. Experimental results on multiple semantic segmentation benchmarks show that our unsupervised segmentation framework specializes in catching semantic representations, which outperforms all the unpretrained and even several pretrained methods. 

Rethinking Alignment and Uniformity in Unsupervised Image Semantic Segmentation

We present a new self-supervised paradigm on point cloud sequence understanding. Inspired by the discriminative and generative self-supervised methods, we design two tasks, namely point cloud sequence based  Contrastive Prediction and Reconstruction (CPR), to collaboratively learn more comprehensive spatiotemporal representations. Specifically, dense point cloud segments are first input into an encoder to extract embeddings. All but the last ones are then aggregated by a context-aware autoregressor to make predictions for the last target segment. Towards the goal of modeling multi-granularity structures, local and global contrastive learning are performed between predictions and targets. To further improve the generalization of representations, the predictions are also utilized to reconstruct raw point cloud sequences by a decoder, where point cloud colorization is employed to discriminate against different frames. By combining classic contrast and reconstruction paradigms, it makes the learned representations with both global discrimination and local perception. We conduct experiments on four point cloud sequence benchmarks, and report the results on action recognition and gesture recognition under multiple experimental settings. The performances are comparable with supervised methods and show powerful transferability.

Contrastive Predictive Autoencoders for Dynamic Point Cloud Self-supervised Learning

Domain generalization in semantic segmentation aims to alleviate the performance degradation on unseen domains through learning domain-invariant features. Existing methods diversify images in the source domain by adding complex or even abnormal textures to reduce the sensitivity to domain-specific features. However, these approaches depends heavily on the richness of the texture bank and training them can be time-consuming. In contrast to importing textures arbitrarily or augmenting styles randomly, we focus on the single source domain itself to achieve the generalization. In this paper, we present a novel adaptive texture filtering mechanism to suppress the influence of texture without using augmentation, thus eliminating the interference of domain-specific features. Further, we design a hierarchical guidance generalization network equipped with structure-guided enhancement modules, which purpose to learn the domain-invariant generalized knowledge. Extensive experiments together with ablation studies on widely-used datasets are conducted to verify the effectiveness of the proposed model, and reveal its superiority over other state-of-the-art alternatives. 

Adaptive Texture Filtering for Single-domain Generalized Segmentation

Considerable progress has recently been made in leveraging CLIP (Contrastive Language-Image Pre-Training) models for text-guided image manipulation. However, all existing works rely on additional generative models to ensure the quality of results, because CLIP alone cannot provide enough guidance information for fine-scale pixel-level changes. In this paper, we introduce CLIPVG, a text-guided image manipulation framework using differentiable vector graphics,  which is also the first CLIP-based general image manipulation framework that does not require any additional generative models. We demonstrate that CLIPVG can not only achieve state-of-art performance in both semantic correctness and synthesis quality, but also is flexible enough to support various applications far beyond the capability of all existing methods.

CLIPVG: Text-Guided Image Manipulation Using Differentiable Vector Graphics

Recent model-extraction attacks on Machine Learning as a Service (MLaaS) systems have moved towards data-free approaches, showing the feasibility of stealing models trained with difficult-to-access data. However, these attacks are ineffective or limited due to the low accuracy of extracted models and the high number of queries to the models under attack. The high query cost makes such techniques infeasible for online MLaaS systems that charge per query.
We create a novel approach to get higher accuracy and query efficiency than prior data-free model extraction techniques. Specifically, we introduce a novel generator training scheme that maximizes the disagreement loss between two clone models that attempt to copy the model under attack. This loss, combined with diversity loss and experience replay, enables the generator to produce better instances to train the clone models. Our evaluation on popular datasets CIFAR-10 and CIFAR-100 shows that our approach improves the final model accuracy by up to 3.42% and 18.48% respectively. The average number of queries required to achieve the accuracy of the prior state of the art is reduced by up to 64.95%. We hope this will promote future work on feasible data-free model extraction and defenses against such attacks.

DisGUIDE: Disagreement-Guided Data-Free Model Extraction

In practice, Wearable Human Activity Recognition (WHAR) models usually face performance degradation on the new user due to user variance. Unsupervised domain adaptation (UDA) becomes the natural solution to cross-user WHAR under annotation scarcity. Existing UDA models usually align samples across domains without differentiation, which ignores the difference among samples. In this paper, we propose an unsupervised domain adaptation model with sample weight learning (SWL-Adapt) for cross-user WHAR. SWL-Adapt calculates sample weights according to the classification loss and domain discrimination loss of each sample with a parameterized network. We introduce the meta-optimization based update rule to learn this network end-to-end, which is guided by meta-classification loss on the selected pseudo-labeled target samples. Therefore, this network can fit a weighting function according to the cross-user WHAR task at hand, which is superior to existing sample differentiation rules fixed for special scenarios. Extensive experiments on three public WHAR datasets demonstrate that SWL-Adapt achieves the state-of-the-art performance on the cross-user WHAR task, outperforming the best baseline by an average of 3.1% and 5.3% in accuracy and macro F1 score, respectively.

SWL-Adapt: An Unsupervised Domain Adaptation Model with Sample Weight Learning for Cross-User Wearable Human Activity Recognition

Deep learning models have achieved tremendous successes in accurate predictions for computer vision, natural language processing and speech recognition applications. However, to succeed in high-risk and safety-critical domains such as healthcare and finance, these deep learning models need to be made reliable and trustworthy. Specifically, they need to be robust and adaptive to real-world environments which can be drastically different from the training settings. In this talk, I will advocate for Bayesian principles to achieve the goal of building robust and adaptive deep learning models. I will introduce a suite of uncertainty quantification methods for Bayesian deep learning, and demonstrate applications enabled by accurate uncertainty estimates, e.g., robust prediction, continual learning and repairing model failures. I will conclude by discussing the research challenges for robust and adaptive deep learning models.

Robust and Adaptive Deep Learning via Bayesian Principles

Vanden Brink H1,2,3*, Burgert TS2, Barral R4,5,6, Gadiraju, M5, Lujan ME1*

1Division of Nutritional Sciences, Cornell University, Ithaca, NY, USA; 
2Division of Endocrinology, Children’s Mercy Hospital, Kansas City, MO, USA; 
3Department of Nutrition, Texas A&M University, College Station, TX, USA; 
4Adolescent Medicine, Children’s Mercy Hospital, Kansas City, MO, USA; 
5University of Missouri Kansas City School of Medicine, Kansas City, MO, USA; 
6 University of Kansas Medical Center, Kansas City, KS, USA

Introduction: Assessment of ovarian features are discouraged in the diagnosis of PCOS in adolescence owing to a lack of normative data and historically poor image quality via transabdominal ultrasonography. However, we have recently shown that ovarian morphology is adequately resolved via transabdominal ultrasound in adolescents and that follicle number and ovarian size reflect circulating AMH within the first gynecological year. Whether ovarian features at this early gynecological stage predict menstrual cycle status at two-years post-menarche (2YR-PM) is unknown. 

Methods: Menstrual records of 19 adolescent females who were evaluated at 6-10 months post-menarche (PM) were collected as a part of a prospective cohort study. Menstrual cycle status was determined at 2YR-PM as regular (three previous inter-menstrual intervals within 21-45 days) or irregular (at least one of the three previous inter-menstrual intervals <21 or >45 days). At 6-10 months PM, participants underwent a fasting blood draw, anthropometry, bioimpedance, and a transabdominal ultrasound of the ovaries. Differences in reproductive and metabolic features at 6-10 months PM were contrasted in those who were regular versus irregular at 2YR-PM via Mann-Whitney U-Tests. Whether reproductive features assessed at 6-10 months PM predicted menstrual cycle status at 2YR-PM was determined via logistic regression analyses.

Results: At 2YR-PM, 6/19 adolescents met the criteria for regular menstrual cycles and 13/19 met the criteria for irregular menstrual cycles. There were no differences in the metabolic profile nor in clinical or biochemical measures of androgens status at 6-10 months PM between those who ultimately achieved regular vs irregular menstrual cycles (P>0.05). However, adolescents with irregular menstrual cycles at 2YR-PM had higher LH concentrations (4.8 [3.5-5.5] vs 3.0 [2.4-3.8] mIU/mL; P=0.025), AMH (5.6 [5.1-7.3] vs 3.0 [1.6-4.5] ng/mL, P=0.023), and tended to have higher numbers of antral follicles 2-5mm in diameter (FNPO 2-5mm) (19 [14-25] vs 10 [7-16], P=0.06) at 6-10 months PM. The likelihood of regular menstrual cycles at 2YR-PM decreased with higher AMH (Odds Ratio 0.49 [0.24-0.98], P=0.01), LH (OR: 0.33 [0.13-1.09], P=0.019) and tended to decrease with greater FNPO 2-5mm (OR: 0.89 [0.76-1.03], P=0.05) at 6-10 months PM. 

Conclusion: Ovarian features and serum LH may serve as early, clinically feasible biomarkers of future persistent menstrual irregularity with defective folliculogenesis preceding metabolic and androgenic derangements  in the timeline to aberrant reproductive maturation. Whether adolescents with persistent menstrual irregularity meet criteria for PCOS will be determined upon conclusion of the cohort study.  

Funding: National Institutes of Health (R21-HD095372), Canadian Institutes of Health Research (Postdoctoral Fellowship #171268). 


Reproductive features captured within the first gynecological year predict persistent menstrual irregularity at two years post-menarche

Background: Thyroid storm is a rare condition but life threatening. Having a high index of suspicion is required especially in patients without previous diagnosis of hyperthyroidism. The increasing use of contrasted iodine studies has caused a resurgence of iodine induced hyperthyroidism which in the most severe cases presents as thyroid storm.
Case Presentation: A 21-year-old female presented with 4 weeks of worsening dyspnea. Associated symptoms included abdominal pain, nausea, vomiting and fever. 3 weeks prior to presentation she underwent CT angiogram of the chest which was negative for pulmonary embolism. She had no past medical history and denied any current medications or recreational drugs. Temperature was 98.4 F, heart rate 107 bpm and blood pressure 125/110. On exam she seemed agitated and anxious. Mild exophthalmos, hyperdynamic precordium, tachycardia, and bilateral pedal edema were also present on exam. Labs showed initial TSH 0.00 uIU/mL, elevated free T4 5.3ng/dL and elevated thyroid receptor antibodies. Pregnancy test was negative. She was diagnosed with thyroid storm and treatment initiated with propranolol, hydrocortisone, methimazole and potassium iodide. Later symptoms improved and she was discharged on propranolol and methimazole with plan to follow-up with Endocrinology.
Discussion: Thyroid storm has multisystem involvement which can mimic other conditions such as sepsis leading to delayed diagnosis. Correct diagnosis requires a complete history and physical examination. Use of the Burch-Wartofsky scale is helpful to support diagnosis of thyroid storm; in the case the Burch-Wartofsky score was 40 suggestive of thyroid storm. Looking for precipitating factors is important. These precipitating factors include infection, myocardial infarction, pregnancy, trauma, and large loads of exogenous iodine (contrast or medications like amiodarone). The phenomenon of hypersecretion of thyroid hormone by autonomous areas within the thyroid gland after an acute iodine load is known as the Jod-Basedow phenomenon. It usually develops within 2 weeks of the iodine load. In patients with symptoms suggestive of thyrotoxicosis, exam findings such as thyromegaly/thyroid bruit, or iodine deficiency, it is suggested to request thyroid hormone assay before an iodine contrasted study. 



The Perfect Storm: Thyrotoxic Crisis Precipitated by Iodinated Contrast

Background: 

Obstructive sleep apnea (OSA) is associated with increased bone mineral density loss. Hypoxia experienced in the OSA population was suspected to stimulate osteoclast activity 

and contribute to osteoporosis. Continuous positive air pressure (CPAP) therapy, the first line treatment for OSA, reduces nocturnal hypoxia. Therefore, this study evaluates the potential for CPAP therapy to slow down the process of bone loss. 

Methods: 

This is a retrospective cohort study reviewing the electronic medical records of patients between the ages of 45 to 70 years old with the diagnosis of OSA and on CPAP therapy. The OSA diagnosis was supported by a home sleep study, and the CPAP compliance was determined by 70% of usage, or an average of more than 4 hours per day with 5 days in a week. We compared the mean difference of T-score per year on Dual-energy X-ray absorptiometry scans in patients compliant with CPAP treatment and patients not compliant with CPAP, utilizing T-test analysis. 

Results: 

There are 113 patients included in the study, with 81 compliant and 32 non-compliant to CPAP therapy. The mean age of the CPAP compliant group (65.7 years) was higher than the non-compliant group (63.2 years). The mean annual change in T-score for lumbar spine, left femur neck, left femur total and left radius were -0.015 ±0.27, -0.003 ±0.19, 0.002 ±0.21, -0.019 ±0.22 respectively in CPAP compliant group, and -0.03 ±0.20, -0.050 ±0.14, -0.014 ±0.17, -0.012 ±0.26 in the non-compliant group. There was no statistical difference in annual bone density changes between the CPAP compliant and non-compliant group. 

Conclusion: 

This is the first study evaluating the effect of CPAP therapy on bone health. The mean reduction in T-score in three out of four areas each year was less in the CPAP compliant group although not reaching statistical significance, possibly limited by the small sample size. Further prospective study and larger sample size might reveal the effect of CPAP on reducing the speed of bone loss in OSA populations.

Underline digital video library