MRISegmentator-Abdomen: A Fully Automated Multi-Organ and Structure Segmentation Tool for T1-weighted Abdominal MRI.

BACKGROUND: Segmentation of organs and structures in abdominal MRI is useful for many clinical applications, such as disease diagnosis and radiotherapy. Current approaches have focused on delineating a limited set of abdominal structures (13 types). To date, there is no publicly available abdominal MRI dataset with voxel-level annotations of multiple organs and structures. Consequently, a segmentation tool for multi-structure segmentation is also unavailable. METHODS: We curated a T1-weighted abdominal MRI dataset consisting of 195 patients who underwent imaging at National Institutes of Health (NIH) Clinical Center. The dataset comprises of axial pre-contrast T1, arterial, venous, and delayed phases for each patient, thereby amounting to a total of 780 series (69,248 2D slices). Each series contains voxel-level annotations of 62 abdominal organs and structures. A 3D nnUNet model, dubbed as MRISegmentator-Abdomen (MRISegmentator in short), was trained on this dataset, and evaluation was conducted on an internal test set and two large external datasets: AMOS22 and Duke Liver. The predicted segmentations were compared against the ground-truth using the Dice Similarity Coefficient (DSC) and Normalized Surface Distance (NSD). FINDINGS: MRISegmentator achieved an average DSC of 0.861$\pm$0.170 and a NSD of 0.924$\pm$0.163 in the internal test set. On the AMOS22 dataset, MRISegmentator attained an average DSC of 0.829$\pm$0.133 and a NSD of 0.908$\pm$0.067. For the Duke Liver dataset, an average DSC of 0.933$\pm$0.015 and a NSD of 0.929$\pm$0.021 was obtained. INTERPRETATION: The proposed MRISegmentator provides automatic, accurate, and robust segmentations of 62 organs and structures in T1-weighted abdominal MRI sequences. The tool has the potential to accelerate research on various clinical topics, such as abnormality detection, radiotherapy, disease classification among others.

Conserved role of unc-79 in ethanol responses in lightweight mutant mice.

The mechanisms by which ethanol and inhaled anesthetics influence the nervous system are poorly understood. Here we describe the positional cloning and characterization of a new mouse mutation isolated in an N-ethyl-N-nitrosourea (ENU) forward mutagenesis screen for animals with enhanced locomotor activity. This allele, Lightweight (Lwt), disrupts the homolog of the Caenorhabditis elegans (C. elegans) unc-79 gene. While Lwt/Lwt homozygotes are perinatal lethal, Lightweight heterozygotes are dramatically hypersensitive to acute ethanol exposure. Experiments in C. elegans demonstrate a conserved hypersensitivity to ethanol in unc-79 mutants and extend this observation to the related unc-80 mutant and nca-1;nca-2 double mutants. Lightweight heterozygotes also exhibit an altered response to the anesthetic isoflurane, reminiscent of unc-79 invertebrate mutant phenotypes. Consistent with our initial mapping results, Lightweight heterozygotes are mildly hyperactive when exposed to a novel environment and are smaller than wild-type animals. In addition, Lightweight heterozygotes exhibit increased food consumption yet have a leaner body composition. Interestingly, Lightweight heterozygotes voluntarily consume more ethanol than wild-type littermates. The acute hypersensitivity to and increased voluntary consumption of ethanol observed in Lightweight heterozygous mice in combination with the observed hypersensitivity to ethanol in C. elegans unc-79, unc-80, and nca-1;nca-2 double mutants suggests a novel conserved pathway that might influence alcohol-related behaviors in humans.

Molecular classification of thyroid nodules using high-dimensionality genomic data.

OBJECTIVE: We set out to develop a molecular test that distinguishes benign and malignant thyroid nodules using fine-needle aspirates (FNA). DESIGN: We used mRNA expression analysis to measure more than 247,186 transcripts in 315 thyroid nodules, comprising multiple subtypes. The data set consisted of 178 retrospective surgical tissues and 137 prospectively collected FNA samples. Two classifiers were trained separately on surgical tissues and FNAs. The performance was evaluated using an independent set of 48 prospective FNA samples, which included 50% with indeterminate cytopathology. RESULTS: Performance of the tissue-trained classifier was markedly lower in FNAs than in tissue. Exploratory analysis pointed to differences in cellular heterogeneity between tissues and FNAs as the likely cause. The classifier trained on FNA samples resulted in increased performance, estimated using both 30-fold cross-validation and an independent test set. On the test set, negative predictive value and specificity were estimated to be 96 and 84%, respectively, suggesting clinical utility in the management of patients considering surgery. Using in silico and in vitro mixing experiments, we demonstrated that even in the presence of 80% dilution with benign background, the classifier can correctly recognize malignancy in the majority of FNA samples. CONCLUSIONS: The FNA-trained classifier was able to classify an independent set of FNAs in which substantial RNA degradation had occurred and in the presence of blood. High tolerance to dilution makes the classifier useful in routine clinical settings where sampling error may be a concern. An ongoing multicenter clinical trial will allow us to validate molecular test performance on a larger independent test set of prospectively collected thyroid FNAs.

A genotype calling algorithm for affymetrix SNP arrays.

MOTIVATION: A classification algorithm, based on a multi-chip, multi-SNP approach is proposed for Affymetrix SNP arrays. Current procedures for calling genotypes on SNP arrays process all the features associated with one chip and one SNP at a time. Using a large training sample where the genotype labels are known, we develop a supervised learning algorithm to obtain more accurate classification results on new data. The method we propose, RLMM, is based on a robustly fitted, linear model and uses the Mahalanobis distance for classification. The chip-to-chip non-biological variance is reduced through normalization. This model-based algorithm captures the similarities across genotype groups and probes, as well as across thousands of SNPs for accurate classification. In this paper, we apply RLMM to Affymetrix 100 K SNP array data, present classification results and compare them with genotype calls obtained from the Affymetrix procedure DM, as well as to the publicly available genotype calls from the HapMap project.

Power calculations for familial aggregation studies.

Family studies are frequently undertaken as the first step in the search for genetic determinants of disease. Significant familial aggregation of disease is suggestive of a genetic etiology for the disease, and may lead to more focused genetic analyses. Many methods have been proposed in the literature for the analysis of family studies. One model that is appealing for its simplicity of computation and the conditional interpretation of its parameters is the quadratic exponential model (e.g., Zhao and Prentice [1990] Biometrika 77:642-648; Betensky and Whittemore [1996] Appl. Stat. 45:422-429; Hudson et al. [2001a] Am. J. Epidemiol. 153:500-514). However, a limiting factor in its application, as well as that of the other proposed methods, is that power and sample size calculations have not been derived. These calculations are essential for investigators who are designing family studies. Here we derive analytic approximations for power for testing for familial aggregation, for both randomly sampled and nonrandomly sampled families. We also present simulation studies of power for both single- and two-disease cases, both under random and nonrandom sampling.

Power calculations for selective genotyping in QTL mapping in backcross mice.

Selective genotyping concerns the genotyping of a portion of individuals chosen on the basis of their phenotypic values. Often individuals are selected for genotyping from the high and low extremes of the phenotypic distribution. This procedure yields savings in cost and time by decreasing the total number of individuals genotyped. Previous work by Darvasi et al. (1993) has shown that the power to detect a QTL by genotyping 40-50 % of a population is roughly equivalent to genotyping the entire sample. However, these power studies have not accounted for different strategies of analysing the data when phenotypes of individuals in the middle are excluded, nor have they investigated the genome-wide type I error rate under these different strategies or different selection percentages. Further, these simulation studies have not considered markers over the entire genome. In this paper, we present simulation studies of power for the maximum likelihood approach to QTL mapping by Lander & Botstein (1989) in the context of selective genotyping. We calculate the power of selectively genotyping the individuals from the middle of the phenotypic distribution when performing QTL mapping over the whole mouse genome.