Our Research

This paper addresses the problem of detecting possible serious bacterial infection (pSBI) of infancy, i.e. a clinical presentation consistent with bacterial sepsis in newborn infants using cranial ultrasound (cUS) images. The captured image set for each patient enables multiview imagery: coronal and sagittal, with geometric overlap. To exploit this geometric relation, we develop a new learning framework, called the intersection-guided Crossview Local-and Image-level Fusion Network (CLIF-Net). Our technique employs two distinct convolutional neural network branches to extract features from coronal and sagittal images with newly developed multi-level fusion blocks. Specifically, we leverage the spatial position of these images to locate the intersecting region. We then identify and enhance the semantic features from this region across multiple levels using cross-attention modules, facilitating the acquisition of mutually beneficial and more representative features from both views. The final enhanced features from the two views are then integrated and projected through the image-level fusion layer, outputting pSBI and non-pSBI class probabilities. We contend that our method of exploiting multi-view cUS images enables a first of its kind, robust 3D representation tailored for pSBI detection. When evaluated on a dataset of 302 cUS scans from Mbale Regional Referral Hospital in Uganda, CLIF-Net demonstrates substantially enhanced performance, surpassing the prevailing state-of-the-art infection detection techniques.

BACKGROUND: Congenital heart disease (CHD) is an important cause of childhood mortality as well as morbidity in children and adults. While genetic risk contributes to the majority of CHD, most individuals with CHD do not have an identified genetic diagnosis. Short tandem repeat (TR) elements are composed of repeated base pair motifs for 2-6 basepairs that are highly polymorphic in length between individuals. These regions had been difficult to study with short read sequencing, and they have not been studied at a large scale in the context of CHD. New software and sequencing platforms have allowed for more accurate TR element genotyping. Therefore, we aimed to identify TR element variants that could impact the expression of known CHD genes.

RESULTS: We identified de novo and inherited TR element variants near known CHD genes in participants with CHD (n = 1,899) in the Pediatric Cardiac Genomics Consortium cohort as well as unaffected participants (n = 1,932) from the Simons Foundation Autism Research Initiative using short-read sequencing followed by variant calling with the gangSTR pipeline. Comparison with long-read sequencing confirmed proband genotypes for 75% (91/120) of the TR element variants identified using short read sequencing. 114 TR element regions had 3 or more de novo TR element variants, compared to an expectation of 74 TR element regions (1.54-fold enrichment, p < 1.5E-5). CHD genes CACNA1C and EVC2 had the strongest enrichment of TR element variants in the CHD cohort, determined by a higher frequency of nearby de novo TR length variants in the CHD cohort compared to the non-CHD cohort. Within CHD trios, there was over-transmission of a TR element variant near Tab 2.

CONCLUSIONS: In a targeted analysis of de novo and transmitted TR element variants in a large cohort of CHD probands, each individual had   1 de novo TR element variant near a CHD gene, and participants with CHD demonstrate clustering of variants within TR element regions. Long-read sequencing confirmed the majority of TR element variants identified using the gangSTR pipeline. De novo variants in known CHD genes were enriched in participants with CHD, with specific enrichment in TR elements near CACNA1C, EVC2, and Tab 2 in the CHD cohort. Many individual TR element variants were in known regulatory regions, but further work is needed to determine their functional impact.

While exome and whole genome sequencing have transformed medicine by elucidating the genetic underpinnings of both rare and common complex disorders, its utility to predict clinical outcomes remains understudied. Here, we use artificial intelligence (AI) technologies to explore the predictive value of whole exome sequencing in forecasting clinical outcomes following surgery for congenital heart defects (CHD). We report results for a prospective observational cohort study of 2,253 CHD patients from the Pediatric Cardiac Genomics Consortium with a broad range of complex heart defects, pre- and post-operative clinical variables and exome sequencing. Damaging genotypes in chromatin-modifying and cilia-related genes are associated with an elevated risk of adverse post-operative outcomes, including mortality, cardiac arrest and prolonged mechanical ventilation. The impact of damaging genotypes is further amplified in the context of specific CHD phenotypes, surgical complexity and extra-cardiac anomalies. The absence of a damaging genotype in chromatin-modifying and cilia-related genes is also informative, reducing the risk for some adverse postoperative outcomes. Thus, genome sequencing enriches the ability to forecast outcomes following congenital cardiac surgery.

OBJECTIVE: We previously identified Paenibacillus species in the cerebrospinal fluid of 44% of infants presenting for neurosurgical evaluation with findings consistent with postinfectious hydrocephalus (PIH) in Eastern Uganda. Here we sought to compare outcomes among hydrocephalic infants with and without Paenibacillus detection at the time of hydrocephalus surgery.

METHODS: In a prospective observational study, 78 infants with PIH who underwent a cerebrospinal fluid (CSF) diversion prior to 90 days of age had a positive CSF polymerase chain reaction result for Paenibacillus species (PP), and 111 had a negative result (PN). The primary outcome was diversion failure-free survival defined as being alive without diversion failure at last patient contact. Secondary outcomes included overall survival and diversion success.

RESULTS: After a median follow-up period of 35.7 months, the primary outcome was observed in 42 PP patients (54%) and in 76 PN patients (68%) (adjusted hazard ratio (aHR), 2.45; 95% confidence interval [CI], 1.42 to 4.22; P=0.001). PP patients who underwent endoscopic diversion had the worst primary event rate (aHR, 6.47; 95% CI, 2.40 to 17.42; P<0.001). Death from any cause occurred in 16 PP patients (20%) and 9 PN patients (8%) (aHR, 3.47; 95% CI, 1.44 to 8.37; P=0.006). Diversion failure occurred in 28 PP patients (36%) and in 29 PN patients (26%) (aHR, 2.24; 95% CI, 1.31 to 3.85; P=0.003).

CONCLUSIONS: In this study, Paenibacillus detection in the CSF at the time of hydrocephalus surgery was associated with a significantly increased rate of the composite of diversion failure or death, death, and diversion failure, and was particularly increased for patients who had an endoscopic diversion.

This paper addresses the problem of detecting possible serious bacterial infection (pSBI) of infancy, i.e. a clinical presentation consistent with bacterial sepsis in newborn infants using cranial ultrasound (cUS) images. The captured image set for each patient enables multiview imagery: coronal and sagittal, with geometric overlap. To exploit this geometric relation, we develop a new learning framework, called the intersection-guided Crossview Local- and Image-level Fusion Network (CLIF-Net). Our technique employs two distinct convolutional neural network branches to extract features from coronal and sagittal images with newly developed multi-level fusion blocks. Specifically, we leverage the spatial position of these images to locate the intersecting region. We then identify and enhance the semantic features from this region across multiple levels using cross-attention modules, facilitating the acquisition of mutually beneficial and more representative features from both views. The final enhanced features from the two views are then integrated and projected through the image-level fusion layer, outputting pSBI and non-pSBI class probabilities. We contend that our method of exploiting multi-view cUS images enables a first of its kind, robust 3D representation tailored for pSBI detection. When evaluated on a dataset of 302 cUS scans from Mbale Regional Referral Hospital in Uganda, CLIF-Net demonstrates substantially enhanced performance, surpassing the prevailing state-of-the-art infection detection techniques.