Accurately determining tumor regions from stereo fluoroscopic images during radiotherapy is a challenging task. As a result, high-density fiducial markers are implanted around tumors in clinical practice as internal surrogates of the tumor, which leads to associated surgical risks. This study was conducted to achieve lung tumor localization without the use of fiducial markers. We propose training a cascade U-net system to perform color to grayscale conversion, enhancement, bone suppression, and tumor detection to determine the precise tumor region. We generated Digitally Reconstructed Radiographs (DRRs) and tumor labels from 4D planning CT images as training data. An improved maximum projection algorithm and a novel color-to-gray conversion algorithm were proposed to improve the quality of the generated training data. Training a bone suppression model using bone-enhanced and bone-suppressed DRRs enables the bone suppression model to achieve better bone suppression performance. The mean peak signal-to-noise ratios in the test sets of the trained translation and bone suppression models are 39.284 ± 0.034 dB and 37.713 ± 0.724 dB, respectively. The results indicate that our proposed markerless tumor localization method is applicable in seven out of ten cases; in applicable cases, the centroid position error of the tumor detection model is less than 1.13 mm; and the calculated tumor center motion trajectories using the proposed network highly coincide with the motion trajectories of implanted fiducial markers in over 60% of captured groups, providing a promising direction for markerless tumor localization tracking methods.
Fiducial marker implantation for tumor localization in radiotherapy is effective but invasive and carries complication risks. To address this, we propose a marker-less tumor tracking framework to explore the feasibility of a cross-patient deep learning model, aiming to eliminate the need for per-patient retraining. A novel degradation model generates realistic simulated data from digitally reconstructed radiographs (DRRs) to train a Restormer network, which transforms clinical fluoroscopic images into clean, DRR-like images. Subsequently, a DUCK-Net model, trained on DRRs, performs tumor segmentation. We conducted a feasibility study using a clinical dataset from 7 lung cancer patients, comprising 100 distinct treatment fields. The framework achieved an average processing time of 179.8 ms per image and demonstrated high accuracy: the median 3D Euclidean tumor center tracking error was mm, with directional errors of mm (LR), mm (SI), and mm (AP). These promising results validate our approach as a proof-of-concept for a cross-patient marker-less tumor tracking solution, though further large-scale validation is required to confirm broad clinical applicability.