A Dempster-Shafer Approach to Trustworthy AI With Application to Fetal Brain MRI Segmentation.

Deep learning models for medical image segmentation can fail unexpectedly and spectacularly for pathological cases and images acquired at different centers than training images, with labeling errors that violate expert knowledge. Such errors undermine the trustworthiness of deep learning models for medical image segmentation. Mechanisms for detecting and correcting such failures are essential for safely translating this technology into clinics and are likely to be a requirement of future regulations on artificial intelligence (AI). In this work, we propose a trustworthy AI theoretical framework and a practical system that can augment any backbone AI system using a fallback method and a fail-safe mechanism based on Dempster-Shafer theory. Our approach relies on an actionable definition of trustworthy AI. Our method automatically discards the voxel-level labeling predicted by the backbone AI that violate expert knowledge and relies on a fallback for those voxels. We demonstrate the effectiveness of the proposed trustworthy AI approach on the largest reported annotated dataset of fetal MRI consisting of 540 manually annotated fetal brain 3D T2w MRIs from 13 centers. Our trustworthy AI method improves the robustness of four backbone AI models for fetal brain MRIs acquired across various centers and for fetuses with various brain abnormalities.

Lung proliferation is dependent on the duration not the timepoint of tracheal occlusion in nitrofen rats with diaphragmatic hernia.

OBJECTIVE: Prenatal tracheal occlusion (TO) promotes lung growth and is applied clinically in fetuses with congenital diaphragmatic hernia (CDH). Limited data are available regarding the effect of duration versus timepoint of TO. Our objective was to document the impact of TO on lung development in the near-term period in rats with nitrofen-induced CDH. METHOD: Nitrofen was administered on embryonic day (ED)9 and fetal TO was performed on ED18.5, 19, or 20 (term = ED22). Sham-operated and untouched littermates served as controls. Lungs were harvested in 0.5-day steps and only fetuses with a left-sided CDH were included in further analyses. Healthy fetuses provided a reference for normal near-term lung development. RESULTS: Duration of TO in the nitrofen rat model for CDH predicts lung growth in terms of lung-body-weight ratio as well as an increased mRNA level of the proliferation marker Ki67. Longer TO also induced a more complex airway architecture. The timepoint of TO was not predictive of lung growth. CONCLUSION: In the nitrofen rat model of CDH, a longer period of TO leads to enhanced lung growth and more refined airway architecture.

Short term post-operative morphing of sacrocolpopexy mesh measured by magnetic resonance imaging.

BACKGROUND: Sacrocolpopexy (SC) involves suspension of the vaginal vault or cervix to the sacrum using a mesh. Following insertion, the meshes have been observed to have undergone dimensional changes. OBJECTIVE: To quantify dimensional changes of meshes following implantation and characterize their morphology in-vivo. DESIGN SETTING AND PARTICIPANTS: 24 patients underwent SC using PolyVinyliDeneFluoride mesh loaded with Fe3O4 particles. Tailored anterior and posterior mesh flaps were sutured to the respective vaginal walls, uniting at the apex. The posterior flap continued to the sacrum and was attached there. Meshes were visualized on magnetic resonance (MR) imaging at 12 [3-12] (median [range]) months postoperatively and 3D models of the mesh were generated. Dynamic MR sequences were acquired during valsalva to record mesh mobility. OUTCOME MEASURES: The area of the vagina effectively supported by the mesh (Effective Support Area (ESA)) was calculated. The 3D models' wall thickness map was analyzed to identify the locations of mesh folding. Intraclass correlation (ICC) was calculated to test the reliability of the methods. To measure the laxity and flatness of the mesh, the curvature and the ellipticity of the sacral flap were calculated. RESULTS: The ESA calculation methodology had ICC = 0.97. A reduction of 75.49 [61.55-78.67] % (median [IQR]) in area, 47.64 [38.07-59.81] % in anterior flap, and of 23.95 [10.96-27.21] % in the posterior flap was measured. The mesh appeared thicker near its attachment at the sacral promontory (n = 19) and near the vaginal apex (n = 22). The laxity of the mesh was 1.13 [1.10-1.16] and 60.55 [49.76-76.25] % of the sacral flap was flat. We could not reliably measure mesh mobility (ICC = 0.16). CONCLUSION: A methodology for complete 3D characterization of SC meshes using MR images was presented. After implantation, the supported area is much lower than what is prepared prior to implantation. We propose this happened during the surgery itself.

Cross-linked xenogenic collagen implantation in the sheep model for vaginal surgery.

The properties of meshes used in reconstructive surgery affect the host response and biomechanical characteristics of the grafted tissue. Whereas durable synthetics induce a chronic inflammation, biological grafts are usually considered as more biocompatible. The location of implantation is another determinant of the host response: the vagina is a different environment with specific function and anatomy. Herein, we evaluated a cross-linked acellular collagen matrix (ACM), pretreated by the anti-calcification procedure ADAPT® in a sheep model for vaginal surgery. Ten sheep were implanted with a cross-linked ACM, and six controls were implanted with a polypropylene (PP; 56 g/m2) control. One implant was inserted in the lower rectovaginal septum, and one was used for abdominal wall defect reconstruction. Grafts were removed after 180 days; all graft-related complications were recorded, and explants underwent bi-axial tensiometry and contractility testing. Half of ACM-implanted animals had palpable induration in the vaginal implantation area, two of these also on the abdominal implant. One animal had a vaginal exposure. Vaginal ACMs were 63 % less stiff compared to abdominal ACM explants (p = 0.01) but comparable to vaginal PP explants. Seven anterior vaginal ACM explants showed areas of graft degradation on histology. There was no overall difference in vaginal contractility. Considering histologic degradation in the anterior vaginal implant as representative for the host, posterior ACM explants of animals with degradation had a 60 % reduced contractility as compared to PP (p = 0.048). Three abdominal implants showed histologic degradation; those were more compliant than non-degraded implants. Vaginal implantation with ACM was associated with graft-related complications (GRCs) and biomechanical properties comparable to PP. Partially degraded ACM had a decreased vaginal contractility.

Host reaction to vaginally inserted collagen containing polypropylene implants in sheep.

OBJECTIVE: We aimed to characterize the effect of vaginal or abdominal mesh insertion and of different collagen augmentation of polypropylene mesh in a sheep model. Outcome measures were passive and active biomechanical properties and semiquantitative morphometry. STUDY DESIGN: Forty-two Texel sheep were used: 6 were nonimplanted controls (n = 6), the rest were implanted with polypropylene mesh (n = 12; Avaulta Solo; Bard Medical, Covington, GA) or collagen-coated meshes: Avaulta Plus (n = 12; Bard Medical) and Ugytex (n = 12; Sofradim International, Trevoux, France). Through a single incision, the rectovaginal septum was dissected and a 35 × 35-mm mesh was sutured to the underlying tissues. Abdominally, a 50 × 50-mm mesh was laid over a primarily sutured full thickness 40-mm longitudinal incisional defect. Animals were explanted after 60 or 180 days (n = 6 per group). Outcome measures were passive biomechanics by biaxial tensiometry, active contractility of vaginal explants, and histologic evidence. RESULTS: Vaginal explants were 2.4 times stiffer compared with native vaginal tissue (P < .001), but without differences in comfort zone stiffness or slope of the load-elongation in the physiologic range between the products that were tested. Collagen coating was associated with a 16-fold reduction in contractile force at 180 days, compared with native vaginal tissue, both for Avaulta Plus (P = .032) and Ugytex (P = .015). Abdominal explants were 1.3-times stiffer compared with native abdominal wall tissue (P < .001) and were 1.9-times stiffer compared with vaginal explants. CONCLUSION: Vaginal mesh implantation yields less stiff explants compared with abdominal explants. Vaginal mesh implantation also alters the passive and active biomechanical properties compared with native vaginal tissues. Collagen matrices did not reduce the number of graft-related complications.