Journal of Clinical and Diagnostic Research. 2018 Feb, Vol-12(2): KC01-KC11 1 1 DOI: 10.7860/JCDR/2018/34311.11217 Original Article Biotechnology Section Intra- and Inter-operator Reproducibility Analysis of Automated Cloud-based Carotid Intima Media Thickness Ultrasound Measurement INTRODUCTION Carotid atherosclerosis leading to stroke is one of the major causes of morbidity in the United States [1,2]. Atherosclerosis disease damages the endothelium and narrows the arteries, hampering oxygenated blood flow [3]. Over time, this blockage can rupture, causing a stroke. To identify the plaque burden in these carotid arteries, ultrasonography examinations are preferred as per the guidelines of American Society of Echocardiography (ASE) [4]. Ultrasound based measurements are: safe, have low acquisition time, provide real-time data, and are fairly economical [5]. Due to harmonic and compound imaging features present in ultrasound (US), high-resolution B-mode grayscale scans can be acquired which allows the visualisation of carotid walls. As a result, it is possible to manually trace the Lumen-Intima (LI) and Media-Adventitia (MA) borders and measure the distance between them, so called cIMT [6]. cIMT has become one of the most widely used biomarkers for risk of stroke and cardiovascular diseases [5,7-14]. Manual cIMT measurements are tedious and prone to errors. Further, these measurements are subject to intra- and inter- observer variability [15]. Several studies affirm the requirement for an automated system for cIMT computation [12,13]. Recent studies have proposed automated cIMT computation methods, but they still suffer from low reproducibility and lack standardisation towards clinical trials [10,15-21] . These automated systems need to ensure the consistency and reliability in their measurements [15,17]. Recently Saba L et al., have proposed an initial design of AtheroCloud, a cloud-based, smart cIMT measurement tool for stroke/cardiovascular risk assessment and risk stratification [22]. The workflow for such a cloud-based automated cIMT measurement system is shown in [Table/Fig-1a]. Retrospective B-mode scans are seldom in non-Digital Imaging and Communications in Medicine (DICOM) format which causes a loss in calibration (or resolution) factors. This puts a constraint on the reliability of the cloud-based automated cIMT measurement system. The novelty of this study is to demonstrate the intra and inter-operator reproducibility for cloud-based automated cIMT measurement system. Further, we hypothesise that even a novice operator can yield high reproducibility when computing cIMT readings. MATERIALS AND METHODS Patient Selection In this study, the data contains a randomly selected set of 100 patients from a pool of 200 patients, acquired from July 2009 to December 2010. The reason for selecting a limited population is to avoid manual tracings on 200 more images which is expensive, tedious, and time-consuming. These patients went through the carotid US for both left and right neck and were retrospectively analysed. The study was institutional review board approved and the images were anonymised. Written informed consent was also provided by all the patients. LUCA SABA 1 , SUMIT K BANCHHOR 2 , TADASHI ARAKI 3 , HARMAN S SURI 4 , NARENDRA D LONDHE 5 , JOHN R LAIRD 6 , KLAUDIJA VISKOVIC 7 , JASJIT S SURI 8 Keywords: AtheroCloud, Atherosclerosis, Stroke, Trial mode ABSTRACT Introduction: Manual carotid Intima Media Thickness (cIMT) measurements are tedious and prone to errors. Further, these measurements are subject to intra and inter-observer variability. Several studies affirm the requirement for an automated system for cIMT computation, but they still suffer from low reproducibility and lack standardisation towards clinical trials. The novelty of this study is to demonstrate the intra and inter- operator reproducibility for a cloud-based automated cIMT measurement system. Aim: To demonstrate the reproducibility analysis and validation of cloud-based automated cIMT measurement systems. Materials and Methods: The reproducibility analysis was performed by two operators at three separate times (six auto readings: 1a, 1b, 1c, 2a, 2b, 2c). For validation of cloud-based cIMT measurement system, we compared the automated readings against the manual readings by the expert. The expert readings were provided by two observers who manually traced the LI/ MA borders at two separate times (four manual readings: 1a, 1b, 2a, 2b). Further, we also performed the variability analysis of the manual readings. Results: The mean Correlation Coefficients (CC) for six intra and nine inter-operator reproducibilities between the auto readings pairs were: 0.99 (p<0.001) and 0.96 (p<0.001), respectively. The mean CCs for two intra and four inter-observer variabilities between the manual readings pairs were 0.94 (p<0.001) and 0.95 (p<0.001), respectively. The accuracy computed between the mean of the six auto readings against each of the four manual readings were: 96.88%, 97.26%, 99.04%, and 98.95%, respectively. While keeping the threshold at 0.9 mm, the ROC using eight combinations give a mean AUC of 0.97±0.01. Conclusion: The proposed cloud-based automated cIMT measurement software system showed high reproducibility. The system can be adapted for routine or clinical (pharmaceutical) trial modes.