A Proposal for an Alternative Approach to Particle Size Method Development During Early-Stage Small Molecule Pharmaceutical Development

Particle size analysis in the pharmaceutical industry has long been a source of debate regarding how best to define measurement accuracy; the degree to which the result of a measurement or calculation conforms to the true value. Defining a “true” value for the size of a particle can be challenging as the output of its measurement will differ because of variations in measurement approaches, instrumental differences and calculation methods. Consequently, for “real” particles, a universal “true” value does not exist and accuracy is therefore not a definable characteristic. Accordingly, precision is then a measure of the ability to reproducibly achieve a measurement of unknown relevance.This article proposes, in place of accuracy, a means to define the “appropriateness” of a measurement in line with the critical quality attributes (CQA) of the material being characterized. The decision as to whether the measurement is correct should involve a link to the CQA; that is, correlation should be demonstrated, without which the measured particle size cannot be defined as a critical material attribute.Correspondingly, methods should also be able to provide sufficient precision to demonstrate discrimination relating to variation in the CQA. The benefits and challenges of this approach are discussed.     

Deep learning with biopsy whole slide images for pretreatment prediction of pathological complete response to neoadjuvant chemotherapy in breast cancer：A multicenter study

IntroductionPredicting pathological complete response (pCR) for patients receiving neoadjuvant chemotherapy (NAC) is crucial in establishing individualized treatment. Whole-slide images (WSIs) of tumor tissues reflect the histopathologic information of the tumor, which is important for therapeutic response effectiveness. In this study, we aimed to investigate whether predictive information for pCR could be detected from WSIs.Materials and methodsWe retrospectively collected data from four cohorts of 874 patients diagnosed with biopsy-proven breast cancer. A deep learning pathological model (DLPM) was constructed to predict pCR using biopsy WSIs in the primary cohort, and it was then validated in three external cohorts. The DLPM could generate a deep learning pathological score (DLPs) for each patient; stromal tumor-infiltrating lymphocytes (TILs) were selected for comparison with DLPs.ResultsThe WSI feature-based DLPM showed good predictive performance with the highest area under the curve (AUC) of 0.72 among the cohorts. Alternatively, the combination of the DLPM and clinical characteristics offered a better prediction performance (AUC >0.70) in all cohorts. We also evaluated the performance of DLPM in three different breast subtypes with the best prediction for the triple-negative breast cancer (TNBC) subtype (AUC: 0.73). Moreover, DLPM combined with clinical characteristics and stromal TILs achieved the highest AUC in the primary cohort (AUC: 0.82) and validation cohort 1 (AUC: 0.80).ConclusionOur study suggested that WSIs integrated with deep learning could potentially predict pCR to NAC in breast cancer. The predictive performance will be improved by combining clinical characteristics. DLPs from DLPM can provide more information compared to stromal TILs for pCR prediction.     

Differentiating autoimmune pancreatitis from pancreatic ductal adenocarcinoma with CT radiomics features

PurposeThe purpose of this study was to determine whether computed tomography (CT)-based machine learning of radiomics features could help distinguish autoimmune pancreatitis (AIP) from pancreatic ductal adenocarcinoma (PDAC).Materials and MethodsEighty-nine patients with AIP (65 men, 24 women; mean age, 59.7 ± 13.9 [SD] years; range: 21–83 years) and 93 patients with PDAC (68 men, 25 women; mean age, 60.1 ± 12.3 [SD] years; range: 36–86 years) were retrospectively included. All patients had dedicated dual-phase pancreatic protocol CT between 2004 and 2018. Thin-slice images (0.75/0.5 mm thickness/increment) were compared with thick-slices images (3 or 5 mm thickness/increment). Pancreatic regions involved by PDAC or AIP (areas of enlargement, altered enhancement, effacement of pancreatic duct) as well as uninvolved parenchyma were segmented as three-dimensional volumes. Four hundred and thirty-one radiomics features were extracted and a random forest was used to distinguish AIP from PDAC. CT data of 60 AIP and 60 PDAC patients were used for training and those of 29 AIP and 33 PDAC independent patients were used for testing.ResultsThe pancreas was diffusely involved in 37 (37/89; 41.6%) patients with AIP and not diffusely in 52 (52/89; 58.4%) patients. Using machine learning, 95.2% (59/62; 95% confidence interval [CI]: 89.8–100%), 83.9% (52:67; 95% CI: 74.7–93.0%) and 77.4% (48/62; 95% CI: 67.0–87.8%) of the 62 test patients were correctly classified as either having PDAC or AIP with thin-slice venous phase, thin-slice arterial phase, and thick-slice venous phase CT, respectively. Three of the 29 patients with AIP (3/29; 10.3%) were incorrectly classified as having PDAC but all 33 patients with PDAC (33/33; 100%) were correctly classified with thin-slice venous phase with 89.7% sensitivity (26/29; 95% CI: 78.6–100%) and 100% specificity (33/33; 95% CI: 93–100%) for the diagnosis of AIP, 95.2% accuracy (59/62; 95% CI: 89.8–100%) and area under the curve of 0.975 (95% CI: 0.936–1.0).ConclusionsRadiomic features help differentiate AIP from PDAC with an overall accuracy of 95.2%.     

Impact of sampling rate on statistical significance for single subject fMRI connectivity analysis

A typical time series in functional magnetic resonance imaging (fMRI) exhibits autocorrelation, that is, the samples of the time series are dependent. In addition, temporal filtering, one of the crucial steps in preprocessing of functional magnetic resonance images, induces its own autocorrelation. While performing connectivity analysis in fMRI, the impact of the autocorrelation is largely ignored. Recently, autocorrelation has been addressed by variance correction approaches, which are sensitive to the sampling rate. In this article, we aim to investigate the impact of the sampling rate on the variance correction approaches. Toward this end, we first derived a generalized expression for the variance of the sample Pearson correlation coefficient (SPCC) in terms of the sampling rate and the filter cutoff frequency, in addition to the autocorrelation and cross‐covariance functions of the time series. Through simulations, we illustrated the importance of the variance correction for a fixed sampling rate. Using the real resting state fMRI data sets, we demonstrated that the data sets with higher sampling rates were more prone to false positives, in agreement with the existing empirical reports. We further demonstrated with single subject results that for the data sets with higher sampling rates, the variance correction strategy restored the integrity of true connectivity.     

Effect of Day Optimizing Throughput (Dot) Knee Software Implementation on Magnetic Resonance Imaging Workflow Efficiency

PurposeAttempts by magnetic resonance (MR) manufacturers to help imaging centres improve patient throughput has led to the development of more automated acquisition. This software is capable of customizing individual scan alignment; potentially improving imaging efficiency and standardizing protocols. However, substantial investments are required to introduce such systems, potentially deterring their widespread application. This study assessed the implementation costs and reduction in examination durations for automated knee MR imaging (MRI) software.Materials and MethodsResearch activities were performed at a community-based academic centre on a 3-Tesla (3-T) system using Siemens' Day Optimizing Throughput (Dot) knee software. Examination acquisition times were extracted from the system before and after software implementation. Fiscal year 2012/13 finances were used to determine the average hourly cost of MRI utilization. Costs associated with automated software implementation were also calculated. Finally, the number of knee scans required to achieve a positive return on investment using the software was established.Results and DiscussionThe mean (standard deviation, sample size) pre- and post-Dot software scan times were 23.20 (4.18, n = 266) and 21.94 (4.51, n = 59) minutes, respectively, for a routine knee scan and 11.88 (1.60, n = 74) and 11.24 (1.51, n = 27) minutes, respectively, for a fast knee scan. The overall weighted average resulted in a 64-second time savings per automated knee examination. This negligible time savings would be extremely difficult to make use of clinically. Dot simplified 29 unique knee protocols to two, improving the consistency of knee examinations. Current Dot software is not compatible with all patients and therefore has limitations that are a concern among MR technologists.ConclusionAdoption of automated knee systems could assist in standardizing protocols; however, the cost of implementation and difficulty in modifying patient scheduling to reflect the minimal time savings would make a financial return unlikely to occur at small- and medium-sized institutions.     

Building and Exploitation of Learning Curves to Train Radiographer Students in X-Ray CT Image Postprocessing

IntroductionThis study aims to construct learning curves related to the realization of standardized postprocessing by radiographer students and to discuss their exploitation and interest.Materials and MethodsThis study was carried out in 21 French students in their 3rd year of training. Two postprocessing protocols in CT (#1 traumatic shoulder; #2 petrous bone) were repeated 15 times by each student. Each achievement was timed to obtain overall learning curves. The realization accuracy was also assessed for each student at each repetition.ResultsThe learning rates for the two protocols are 63% and 56%, respectively. The number of repetitions to reach the reference time for each protocol is 11 and 12, respectively. In both protocols, the standard deviations are significantly reduced and stabilized during repetitions. The mean accuracy progresses more quickly in protocol #1.DiscussionThe measured learning rates reflect a rapid learning process for each protocol. The analysis of the standard deviations shows that students have reached a homogeneous level. The average times and accuracies measured during the last repetitions show that the group has reached a high level of performance. Building learning curves helps students measure their progress and motivates them.ConclusionObtaining learning curves allows trainers/supervisors to qualify the learning difficulty of a task while motivating students/radiographers. The use of learning curves is inline with the competency-based training paradigm.