Search for: All records

Deep learning (DL) models have demonstrated state-of-the-art performance in the classification of diagnostic imaging in oncology. However, DL models for medical images can be compromised by adversarial images, where pixel values of input images are manipulated to deceive the DL model. To address this limitation, our study investigates the detectability of adversarial images in oncology using multiple detection schemes. Experiments were conducted on thoracic computed tomography (CT) scans, mammography, and brain magnetic resonance imaging (MRI). For each dataset we trained a convolutional neural network to classify the presence or absence of malignancy. We trained five DL and machine learning (ML)-based detection models and tested their performance in detecting adversarial images. Adversarial images generated using projected gradient descent (PGD) with a perturbation size of 0.004 were detected by the ResNet detection model with an accuracy of 100% for CT, 100% for mammogram, and 90.0% for MRI. Overall, adversarial images were detected with high accuracy in settings where adversarial perturbation was above set thresholds. Adversarial detection should be considered alongside adversarial training as a defense technique to protect DL models for cancer imaging classification from the threat of adversarial images. 

Despite considerable advances in recent years, challenges in delivery and storage of biological drugs persist and may delay or prohibit their clinical application. Though nanoparticle-based approaches for small molecule drug encapsulation are mature, encapsulation of proteins remains problematic due to destabilization of the protein. Reverse micelles composed of decylmonoacyl glycerol (10MAG) and lauryldimethylamino-N-oxide (LDAO) in low-viscosity alkanes have been shown to preserve the structure and stability of a wide range of biological macromolecules. Here, we present a first step on developing this system as a future platform for storage and delivery of biological drugs by replacing the non-biocompatible alkane solvent with solvents currently used in small molecule delivery systems. Using a novel screening approach, we performed a comprehensive evaluation of the 10MAG/LDAO system using two preparation methods across seven biocompatible solvents with analysis of toxicity and encapsulation efficiency for each solvent. By using an inexpensive hydrophilic small molecule to test a wide range of conditions, we identify optimal solvent properties for further development. We validate the predictions from this screen with preliminary protein encapsulation tests. The insight provided lays the foundation for further development of this system toward long-term room-temperature storage of biologics or toward water-in-oil-in-water biologic delivery systems. 

In recent years, it has increasingly been recognized that spatial visualization skills are important in supporting student success in Science, Technology, Engineering, and Math (STEM) education and retention of these students in STEM careers. Many first year college engineering programs and high schools with pre-engineering curriculum have incorporated spatial visualization training into their courses. However, there is no reason why spatial visualization training could not occur at a much younger age, like elementary school. Often at the high school and college level, the Purdue Spatial Visualization Test: Rotations (PSVT:R), which is recognized as a gold standard assessment tool, is used to measure students’ spatial skills learning gains. The PSVT:R is a 20 minute timed test consisting of 30 three-dimensional rotations problems. While the PSVT:R test has been well validated, there are benefits to developing alternative methods of assessing spatial visualization skills suitable for elementary school grades. Researchers at XXX developed an assembly pre- and post- test based upon a timed Lego™ exercise. Students are timed to see how long it would take them to build small Lego shapes using only a picture of the final assembly, but no instructions. The test was implemented at the beginning and then at the end of the quarter/semester. The beauty of this assessment is that it lends itself well to K-12 students. The 20 minute, timed PSVT:R test is too challenging for elementary aged children and is not engaging. In order to validate the new instrument, the Lego™ Assembly test was implemented in a pilot study in 2 college freshman engineering courses using students who could do both the PSVT:R and the Lego Assembly™ assessments. At the beginning of the class all students took the PSVT:R test, and half the students performed a Lego™ assembly of one shape and the other half did the assembly test with another shape. During the class the students completed spatial visualization training which taught them how to sketch orthographic and isometric assignments using a mobile sketching app. At the end of the class the PSVT:R test was repeated for all students. The Lego Assembly™ test was also completed, but the students switched which shape they were building. This approach allowed us to normalize the difficulty of the assembly tasks based upon the average time it took to build the shapes in the pre-test. The Lego Assembly™ test was first implemented in winter and spring of 2018. The data showed a correlation of the R-Squared of 0.11 between the assembly times and the PSVT:R scores in pre-test and 0.14 in post-test. However, analysis of the assembly times indicated that the difficulty of the 2 Lego shapes were significantly different, which could skew the normalization of the assembly times. Accordingly, the test was repeated in winter and spring 2019 with Lego shapes of similar difficulty. This paper describes the results of the assembly tess in all 4 classes, and its suitability for a spatial visualization assessment for elementary school age students. 

Effective insect pollination requires appropriate responses to internal and external environmental cues in both the plant and the pollinator.Helianthus annuus, a highly outcrossing species, is marked for its uniform eastward orientation of mature pseudanthia, or capitula. Here we investigate how this orientation affects floral microclimate and the consequent effects on plant and pollinator interactions and reproductive fitness.

We artificially manipulated sunflower capitulum orientation and temperature in both field and controlled conditions and assessed flower physiology, pollinator visits, seed traits and siring success.

East‐facing capitula were found to have earlier style elongation, pollen presentation and pollinator visits compared with capitula manipulated to face west. East‐facing capitula also sired more offspring than west‐facing capitula and under some conditions produced heavier and better‐filled seeds. Local ambient temperature change on the capitulum was found to be a key factor regulating the timing of style elongation, pollen emergence and pollinator visits.

These results indicate that eastward capitulum orientation helps to control daily rhythms in floral temperature, with direct consequences on the timing of style elongation and pollen emergence, pollinator visitation, and plant fitness.