Development of DICOM-based computer-aided system and conformity index for evaluation of intensity modulated radiation therapy plans for head-and-neck cancer

Abstract

Introduction: Intensity modulated radiation therapy (IMRT) has gained popularity in the treatment of cancers. Due to its complex anatomy, manual evaluation of IMRT plans for head-and-neck cancer has been especially challenging necessitating efficient and objective assessment tools. Thus" the aim of this study was to develop a computer-aided evaluation (CAR) system and a personalized target conformity index (Cl) for automatic evaluation of IMRT plans for head-and-neck cancer. Methodology: The CAE and a personalized target CI were developed based on the four-phase spiral model. Utilizing the MATLAB program language, a set of routines was written to parse key data from the Digital Imaging and Communications in Medicine for radiation therapy (DICOM RT) objects. After data reconstruction, an algorithm for detection of violations was developed to extract the overdose and underdose regions. Thirty IMRT plans for nasopharyngeal carcinoma were collected for evaluation of CAE, system performance. Normalized root mean square deviation (NRMSD) was computed for comparison between manually extracted and CAE-extracted data. To assess the plan quality discerning power of the personalized target CI, three computed tomography data sets of nasopharyngeal carcinoma patientswere collected. Ten IMRT plans were generated from each data set They were ranked and compared with different conformity indices. The coefficient of variance was calculated for each data set to compare the degree of variation from personalized target CI to other existing indices. Mixed-design analysis of variance (ANOVA) was performed to explore the impact of use of CAE system with personalized target CIs on plan evaluation time. Results: The CAE and a personalized target CI for automatic evaluation of IMRT plans for head-and-neck cancer was developed. Three major graphical user interfaces (GUIs) were created to intelligently lead the planners through the steps of plan evaluation process. The CAE-computed dose volume histogram (DVH) results were in good agreement with the manually extracted data with a NRMSD of less than 0.05%. With the aid of the CAE system, the per-region detection performance for small-size spots was greatly improved by 2.36-fold and 3.99-fold for experienced and inexperienced planners, respectively. Compared with other commonly used indices, the personalized target CIs resulted in the largest coefficient of variance among 10 IMRT plans for each data set, indicating that its discerning power was the best among the indices being compared. The ANOVA results indicated that the evaluation time with the aid of CAE system was significantly shortened by 1.88-fold than that without using the CAE system, regardless of the level of experience. Conclusion: The CAE system with personalized target CI demonstrated good applicability of DICOM RT objects in data mining. The use of CAE system with personalized target CI could eliminate human errors, provide plan quality control and enhance efficiency in evaluating IMRT plans for head-and-neck cancer. By taking individual tumor geometry into account, the superiority of personalized target CI in plan discerning power was demonstrated. As an effective data mining tool, the CAE system with personalized target CI could be adopted in the evaluation of treatment plans other than IMRT.