||Quality Assurance of Lung Cancer CT Screening CAD Systems : Feasibility Study with PSF-based Virtual Nodules
肺がんＣＴ検診コンピュータ支援診断システムの品質保証 : ＰＳＦに基づく仮想結節を用いた可能性研究
Marasinghe Arachchige Janaka Chandrasiri Marasinghe
75 , 2015-03-23 , 新潟大学
学位の種類: 博士（保健学）. 報告番号: 甲第4057号. 学位記番号: 新大院博（保）甲第14号. 学位授与年月日: 平成27年3月23日
Computer-aided detection/diagnosis (CAD) is increasingly used in clinical practice. CAD systems are useful for decision making in detection and interpretation of diseases by clinicians. Therefore, quality assurance (QA) for CAD has a key role in clinics and it helps end users to make aware of changes in CAD performance both due to intentional or unintentional causes. However, there are no QA requirements for CAD in clinical use at present. CAD subcommittee of American Association of Physicists in Medicine (AAPM) has recently published the general guidelines for all CAD systems used in clinical practice. Further to that need of research studies to find appropriate methods for QA of CAD systems has been pointed out by the publication. Purpose of my study is to introduce a new method to assess lung cancer CT screening CAD systems’ performance effectively and perform QA of the same CAD system. First part of the study was performed to verify the point spread function (PSF)-based virtual nodules. 19 clinical lung nodules of 18 subjects were used from screening data set of a general hospital for the verification study. These nodules were confirmed by experienced lung screening radiologist. PSF-based nodules were simulated with quite similar size and density of each clinical nodule. Then superimposed on same images and CAD were performed. Similar free response receiver characteristic (FROC) curves were obtained for both clinical lung nodules and comparable PSF-based virtual nodules. This result implies that the CAD system has shown a similar performance on both clinical nodules and PSF-based virtual nodules. Therefore, this result verifies the PSF-based virtual nodules. Then the feasibility of PSF-based virtual nodules was checked for assessing CAD system performance in detail. PSF-based nodules were simulated with various sizes and densities and superimposed on clinical images to assess the CAD performance dependence on nodule size and density and slice thickness. FROC curves are used to analyze results. Nodule size study shows true positive fraction (TPF) is higher for larger nodules. Nodule density study shows that TPF is larger for higher density nodules. Slice thickness study shows that higher percentages of nodules on thick slices were detected by the tested CAD system. Therefore, tested lung cancer CT screening CAD system shows tendency of performance on high dense large nodules on thick slices. Similarly the proposed PSF-based virtual nodules can be used to assess the performance of a CAD system with images from various clinics with different scan/reconstruction conditions. PSF-based virtual nodule method applies its own images of a clinic which includes same scan/reconstruction parameters and avoids the CAD performance dependence on special resolution conditions. Therefore, this method is feasible to apply for QA of lung cancer CT screening CAD systems at any different clinical sited for different CAD systems. Study results show that practical implementation of QA for CAD system is possible with the proposed method. Therefore, this study proposed PSF-based virtual nodules based QA protocol for lung cancer CT screening CAD system.