Logiq e9

We evaluated the performance of radiologists and deep learning algorithm for distinguishing benign from malignant nodules.
The performance of radiologists was calculated based on their assignment of the likelihood of malignancy which were encoded as ordinal numbers. The performance of deep learning was calculated based on the likelihood of malignancy returned by the network. We used area under the receiver operating characteristic curve (AUC) as the performance metric. The calculation of AUC was achieved using pROC package in R.
Confidence intervals were estimated using bootstrap [14] with 2000 stratified bootstrap replicates. Statistical comparison between the performances of radiologists and the performance of deep learning was conducted using DeLong Method [15] In order to assess the impact of ultrasound scanner manufacturer and model, we repeated this analysis for three individual scanner types: ATL HDI 5000, GE LOGIQ E9, and Philips iU22 which together accounted for 95.5% of all images.

Our team previously developed a deep learning algorithm for risk stratification of thyroid nodules on ultrasound [6] . Briefly, the model was based on 1631 nodules from another institution and were acquired from several manufacturers: 94.4% from Siemens Antares and Elegra, 4.0% from GE LOGIQ E9, 0.9% from Philips iU22, and 0.4% from Philips ATL HDI 5000. No nodules from our own institution were used in initial model Then, 50% dropout [13] was added for regularization in the training phase. Finally, one fully connected layer with one output is assigned and sigmoid function is used. The final output of the network was the probability of malignancy.

Institutional review board approval was obtained for a retrospective review of medical records between Jan 1, 2001 to December 31, 2011, revealing 130 women with early pregnancy loss and subsequent diagnosis of molar pregnancy diagnosed during routine histopathologic examination of gestational products. Of these, 70 had ultrasound images available in the radiology archive and were included in this study.
Clinical data including patient age, gravidity and parity, quantitative beta-hCG levels, gestational age (GA) by last menstrual period (LMP), and histopathologic diagnosis were obtained from the medical records. Ultrasound imaging was performed on a GE Logiq 9, GE Logiq E9, or Philips iU22 and included transabdominal (3-8 MHz) and transvaginal (5-10 MHz) pelvic imaging with grey scale and cine-capture series on all patients. Color and spectral Doppler images were obtained when desired by the sonographer and/or interpreting sonologist in order to further characterize the findings on greyscale