Code is seldom written in a single left-to-right pass and is instead repeatedly edited and refined. We introduce INCODER, a unified generative model that can perform program synthesis (via left-to-right generation) as well as editing (via infilling). InCoder is trained to generate code files from a large corpus of permissively licensed code, where regions of code have been randomly masked and moved to the end of each file, allowing code infilling with bidirectional context. Our model is the first large generative code model that is able to infill arbitrary regions of code, which we evaluate in a zero-shot setting on challenging tasks such as type inference, comment generation, and variable re-naming. We find that the ability to condition on bidirectional context substantially improves performance on these tasks, while still performing comparably on standard program synthesis benchmarks in comparison to left-to-right only models pretrained at similar scale. The INCODER models and code are publicly released.2 * Equal contribution 2 https://sites.google.com/view/incoder-code-models

A retrospective study was carried out to evaluate the diagnostic accuracy and the factors influencing the diagnostic accuracy of 648 procedures of ultrasound-guided percutaneous needle biopsy (PNB) for peripheral pulmonary lesions (PPLs). We reviewed the histopathology results, the clinical records and the procedure reports of these 648 biopsies and the final diagnoses of 637 PPLs to determine the diagnostic accuracy of ultrasound-guided PNB. Factors that influenced the diagnostic accuracy were assessed by analysis of the biopsy procedures, which were classified as diagnostic cases (true-positive and true-negative) and non-diagnostic cases (false-positive, false-negative and indeterminate). Statistical analyses of factors that related to patient demographic characteristics, lesion characteristics and biopsy details were performed to determine possible effects on diagnostic accuracy. Biopsies were successfully performed in all cases, and 11 patients underwent second biopsies for the same lesions. Among the 637 PPLs, there were 326 (51.2%) malignant lesions, 272 (42.7%) benign lesions and 39 (6.1%) indeterminate lesions. Of the 272 benign lesions, 114 (41.9%) were found to be tuberculous. The overall diagnostic accuracy was 81.8%, and the rates of hemoptysis, symptomatic pneumothorax and chest-tube insertion were 8.0%, 1.7% and 0.9%, respectively. Lesions sizes were divided into 3 groups according to the measurement by ultrasound. For lesions that measured ≤20 mm, 21-49 mm and ≥50 mm, the diagnostic accuracy was 72.0%, 86.8% and 79.7%, while sensitivity and specificity were 54.3%-79.2%, 88.3%-90.7% and 79.4%-89.5% and 77.3%-100%, 96.8%-100% and 58.6%-100%, respectively. Diagnostic accuracy was significantly affected by lesion size when lesion size was measured by ultrasound (p = 0.006) and computed tomography (CT) (p = 0.001). In the 3 lesion groups of ≤20 mm, 21-49 mm or ≥50 mm, diagnostic accuracy among each group was significantly different (p <0.001). When lesion size was measured by ultrasound (p <0.001) and CT (p <0.001) and the 3 groups were analyzed (p <0.001), there was a statistically significant relationship between lesion size and the presence of necrosis. The rates of the presence of necrosis in lesions that measured ≤20 mm, 21-49 mm and ≥50 mm were 3.9%, 11.7% and 28.8%, respectively. No significance was found for age (p = 0.119), gender (p = 0.25), lesion location (p = 0.55), the presence of necrosis (p = 0.226), patient position (p = 0.25), needle size (p = 0.26), puncture angle (p = 0.34) and needle passes (p = 0.21). Ultrasound-guided PNB is an effective and safe diagnostic method for PPLs; the diagnostic accuracy is significantly affected by lesion size and decreases in smaller (≤20 mm) and larger (≥50 mm) lesions.

No abstract

No abstract