Atomist or holist? A diagnosis and vision for more productive interdisciplinary AI ethics dialogue.

In response to growing recognition of the social impacts of new artificial intelligence (AI)-based technologies, major AI and machine learning (ML) conferences and journals now encourage or require papers to include ethics impact statements and undergo ethics reviews. This move has sparked heated debate concerning the role of ethics in AI research, at times devolving into name calling and threats of "cancellation." We diagnose this conflict as one between "atomist" and "holist" ideologies. Among other things, atomists believe facts are and should be kept separate from values, while holists believe facts and values are and should be inextricable from one another. With the goal of reducing disciplinary polarization, we draw on numerous philosophical and historical sources to describe each ideology's core beliefs and assumptions. Finally, we call on atomists and holists within the ever-expanding data science community to exhibit greater empathy during ethical disagreements and propose four targeted strategies to ensure AI research benefits society.

Reliability of radiographic measurements for hip dysplasia using a reduced radiation technique.

The purpose of this study was to examine whether measurement accuracy is impacted with a low radiation radiograph technique. Treatment decisions for hip dysplasia require serial radiographs to monitor acetabular anatomy. However, radiation use creates concern for future malignancy. Standard radiograph technique utilizes a grid to enhance clarity, but requires a larger X-ray dose. We hypothesized grid removal would lower radiation exposure without decreasing measurement accuracy. This is a retrospective cohort study of patients with idiopathic DDH and neuromuscular hip displacement at a single institution from 2019 to 2020 receiving surveillance radiographs. A cohort receiving modified technique radiographs (grid removed) was compared to an age-matched cohort receiving standard technique radiographs. Four examiners measured radiographs on two separate occasions, a minimum of 2 weeks apart. Migration percentage (MP) was measured for neuromuscular patients and acetabular index (AI) for idiopathic DDH. Intraclass correlation coefficient (ICC) with a 95% confidence interval (CI) was used to evaluate inter-observer and intra-observer agreement for measurements. Average radiation dose for both techniques. A total of 134 age-matched patients were included for measurement of MP. Thirty age-matched patients were included for AI measurements. Inter and intra-rater reliability of MP and AI were not different with or without use of the grid. There was a 50-75% decrease in radiation dose per radiograph following grid removal, dependent on patient size. The adjusted radiograph technique showed no significant difference in MP or AI reliability. A slight decrease in picture contrast did not jeopardize measurement reliability and decreased radiation exposure for the patient significantly. This simple technique can be useful for patients requiring multiple radiographs over time.

Adjusting to the GDPR: The Impact on Data Scientists and Behavioral Researchers.

Rapid growth in the availability of behavioral big data (BBD) has outpaced the speed of updates to ethical research codes and regulation of data privacy and human subjects' data collection, storage, and use. The introduction of the European Union's (EU's) General Data Protection Regulation (GDPR) in May 2018 will have far-reaching effects on data scientists and researchers who use BBD, not only in the EU, but around the world. Consequently, many companies are struggling to comply with the Regulation. At the same time, academics interested in research collaborations with companies are finding it more difficult to obtain data. In light of the importance of BBD in both industry and academia, data scientists and behavioral researchers would benefit from a deeper understanding of the GDPR's key concepts, definitions, and principles, especially as they apply to the data science workflow. We identify key GDPR concepts and principles and describe how they can impact the work of data scientists and researchers in this new data privacy regulation era.

Evaluation of cassette-based digital radiography detectors using standardized image quality metrics: AAPM TG-150 Draft Image Detector Tests.

The purpose of this study was to evaluate several of the standardized image quality metrics proposed by the American Association of Physics in Medicine (AAPM) Task Group 150. The task group suggested region-of-interest (ROI)-based techniques to measure nonuniformity, minimum signal-to-noise ratio (SNR), number of anomalous pixels, and modulation transfer function (MTF). This study evaluated the effects of ROI size and layout on the image metrics by using four different ROI sets, assessed result uncertainty by repeating measurements, and compared results with two commercially available quality control tools, namely the Carestream DIRECTVIEW Total Quality Tool (TQT) and the GE Healthcare Quality Assurance Process (QAP). Seven Carestream DRX-1C (CsI) detectors on mobile DR systems and four GE FlashPad detectors in radiographic rooms were tested. Images were analyzed using MATLAB software that had been previously validated and reported. Our values for signal and SNR nonuniformity and MTF agree with values published by other investigators. Our results show that ROI size affects nonuniformity and minimum SNR measurements, but not detection of anomalous pixels. Exposure geometry affects all tested image metrics except for the MTF. TG-150 metrics in general agree with the TQT, but agree with the QAP only for local and global signal nonuniformity. The difference in SNR nonuniformity and MTF values between the TG-150 and QAP may be explained by differences in the calculation of noise and acquisition beam quality, respectively. TG-150's SNR nonuniformity metrics are also more sensitive to detector nonuniformity compared to the QAP. Our results suggest that fixed ROI size should be used for consistency because nonuniformity metrics depend on ROI size. Ideally, detector tests should be performed at the exact calibration position. If not feasible, a baseline should be established from the mean of several repeated measurements. Our study indicates that the TG-150 tests can be used as an independent standardized procedure for detector performance assessment.

Evaluation of techniques for slice sensitivity profile measurement and analysis.

The purpose of this study was to compare the resulting full width at half maximum of slice sensitivity profiles (SSP) generated by several commercially available point response phantoms, and determine an appropriate imaging technique and analysis method. Four CT phantoms containing point response objects designed to produce a delta impulse signal used in this study: a Fluke CT-SSP phantom, a Gammex 464, a CatPhan 600, and a Kagaku Micro Disc phantom. Each phantom was imaged using 120 kVp, 325 mAs, head scan field of view, 32 × 0.625 mm helical scan with a 20 mm beam width and a pitch of 0.969. The acquired images were then reconstructed into all available slice thicknesses (0.625 mm - 5.0 mm). A computer program was developed to analyze the images of each dataset for generating a SSP from which the full width at half maximum (FWHM) was determined. Two methods for generating SSPs were evaluated and compared by choosing the mean vs. maximum value in the ROI, along with two methods for evaluating the FWHM of the SSP, linear interpolation and Gaussian curve fitting. FWHMs were compared with the manufacturer's specifications using percent error and z-test with a significance value of p < 0.05. The FWHMs from each phantom were not significantly different (p ≥ 0.089) with an average error of 3.5%. The FWHMs from SSPs generated from the mean value were statistically different (p ≤ 3.99 × 10¹³). The FWHMs from the different FWHM methods were not statistically different (p ≤ 0.499). Evaluation of the SSP is dependent on the ROI value used. The maximum value from the ROI should be used to generate the SSP whenever possible. SSP measurement is independent of the phantoms used in this study.