Non-invasive imaging modalities used in the diagnosis of lung cancer, such as Positron Emission Tomography (PET) or Computed Tomography (CT), currently provide insuffcient information about the cellular make-up of the lesion microenvironment, unless they are compared against the gold standard of histopathology.The aim of this retrospective study was to build a robust imaging framework for registering in vivo and post-operative scans from lung cancer patients, in order to have a global, pathology-validated multimodality map of the tumour and its surroundings.Initial experiments were performed on tissue-mimicking phantoms, to test different shape reconstruction methods. The choice of interpolator and slice thickness were found to affect the algorithm's output, in terms of overall volume and local feature recovery. In the second phase of the study, nine lung cancer patients referred for radical lobectomy were recruited. Resected specimens were inflated with agar, sliced at 5 mm intervals, and each cross-section was photographed. The tumour area was delineated on the block-face pathology images and on the preoperative PET/CT scans.Airway segments were also added to the reconstructed models, to act as anatomical fiducials. Binary shapes were pre-registered by aligning their minimal bounding box axes, and subsequently transformed using rigid registration. In addition, histopathology slides were matched to the block-face photographs using moving least squares algorithm.A two-step validation process was used to evaluate the performance of the proposed method against manual registration carried out by experienced consultants. In two out of three cases, experts rated the results generated by the algorithm as the best output, suggesting that the developed framework outperforms the current standard practice.
|Date of Award||4 Jun 2020|
- University Of Strathclyde
|Supervisor||Stephen Marshall (Supervisor) & Jinchang Ren (Supervisor)|