Tracking tumor boundary in MV-EPID images without implanted markers: A feasibility study

Current radiation therapy protocols often struggle to maintain precise tumor targeting due to internal organ motion. Clinicians frequently rely on invasive markers to monitor target positions during treatment delivery. This reliance introduces unnecessary patient discomfort and potential procedural risks. No prior work had resolved the challenge of tracking soft tissue boundaries without these artificial aids. Existing imaging systems generate significant noise, complicating automated detection of moving targets. That uncertainty drove the need for robust computational solutions capable of handling low-contrast environments. Prior research has shown that traditional manual contouring is too slow for real-time clinical workflows. This gap motivated the development of automated, markerless tracking strategies for megavoltage imaging systems.

The primary aim is to develop a markerless tracking algorithm for monitoring tumor boundaries in megavoltage electronic portal imaging device sequences. This research addresses the need for non-invasive methods to track targets during image-guided radiation therapy. The authors seek to eliminate the requirement for implanted markers, which often complicate clinical procedures. They focus on creating a robust mathematical model capable of automatic boundary detection. The study investigates whether a level set method can accurately follow tumor motion in real-time. This effort is motivated by the desire to improve targeting precision during patient irradiation. The researchers intend to demonstrate the feasibility of this approach using both phantom and clinical data. They aim to provide a foundation for future applications in adaptive beam delivery and dose evaluation.

The investigators designed a level set method-based algorithm to monitor target movement within image sequences. Their review approach involved testing the software on three distinct categories of imaging data. They first applied the tool to a four-dimensional phantom sequence to establish baseline performance. Next, the team utilized four digitally deformable phantom sequences characterized by varying levels of image noise. They also incorporated four clinical sequences obtained during actual lung cancer treatment sessions. The researchers evaluated accuracy by calculating the centroid localization error and volume overlap index. They compared their results against two existing tracking techniques to assess relative performance improvements. This systematic evaluation confirmed the utility of the proposed mathematical framework for clinical applications.

The algorithm achieved a centroid localization error of 0.23 millimeters and a volume overlap index of 95.6 percent for the four-dimensional phantom. For the digital phantom sequences, the total error was 0.11 millimeters with a 96.7 percent overlap index. The clinical sequences yielded a centroid localization error of 0.32 millimeters and a 72.1 percent volume overlap index. These findings demonstrate the effectiveness of the proposed method in both localization and boundary monitoring. The authors report that their technique outperforms two previously established tracking algorithms in terms of localization accuracy. The data indicate that the system successfully identifies tumor targets without requiring implanted markers. The results highlight the potential for high-precision tracking across different types of imaging datasets. This performance suggests that the model is suitable for identifying visible targets in real-time.

The authors propose that their mathematical framework effectively monitors tumor positions in real-time. This approach demonstrates superior localization precision compared to two alternative tracking methods. The study confirms that the algorithm functions reliably across both simulated phantoms and actual patient datasets. Researchers suggest that providing continuous boundary data could enhance future adaptive beam delivery systems. The team indicates that their technique supports accurate dose evaluation during active irradiation sessions. These findings imply that markerless tracking is a viable strategy for improving therapeutic outcomes. The authors conclude that their model maintains high performance even when processing clinical images with varying noise levels. This work provides a foundation for integrating automated boundary detection into standard radiotherapy clinical practice.