Master of Engineering Science
Musculoskeletal Health Research
The quality of patient care associated with diagnostic radiology is proportionate to a physician's workload. Segmentation is a fundamental limiting precursor to diagnostic and therapeutic procedures. Advances in machine learning aims to increase diagnostic efficiency to replace single applications with generalized algorithms. We approached segmentation as a multitask shape regression problem, simultaneously predicting coordinates on an object's contour while jointly capturing global shape information. Shape regression models inherent point correlations to recover ambiguous boundaries not supported by clear edges and region homogeneity. Its capabilities was investigated using multi-output support vector regression (MSVR) on head and neck (HaN) CT images. Subsequently, we incorporated multiplane and multimodality spinal images and presented the first deep learning multiapplication framework for shape regression, the holistic multitask regression network (HMR-Net). MSVR and HMR-Net's performance were comparable or superior to state-of-the-art algorithms. Multiapplication frameworks bridges any technical knowledge gaps and increases workflow efficiency.
Summary for Lay Audience
The development of new image analysis techniques has allowed doctors to better understand the content of an image. Segmentation, a technique to isolate regions of interest, is used in medical interventions such as disease detection, tracking disease progression, and evaluating for surgical procedures, and radiation therapy. The integration of artificial intelligence (AI) technology into medical image analysis can enhance and streamline image segmentation practices. However, existing methods are organ-specific and cannot be adapted. Multitask learning techniques were explored to create a machine learning framework for multiple applications. We approached the segmentation task as a multitask prediction problem (estimating multiple tasks such as organ class, location, and boundary points simultaneously) and introduced shape or boundary point regression; an innovative new technique. Shape regression directly predicts coordinates of an object's shape contour and simultaneously captures its shape. Compared to conventional image pixel segmentation, shape regression can model the natural correlation between points to recover unclear boundaries not supported by clear edges and uniform pixel regions. To determine if a generalized algorithm using shape regression was feasible, we first implemented the technique using a traditional machine learning method, multi-output support vector regression (MSVR). MSVR was applied to head and neck (HaN) CT images consisting of 18 target organs. In another study, we used a more modern machine learning technique, deep learning, for shape regression and expanded the application scope for MR and CT spinal images in both axial and sagittal planes. We presented the first deep learning multiapplication framework for shape regression, the holistic multitask regression network (HMR-Net). MSVR and HMR-Net's performance were similar or superior to current algorithms in literature. The successful performance of an automated multiapplication framework provides many benefits to clinical routine. It bridges any technical knowledge gaps and increases workflow efficiency.
Tam, Clara, "Machine Learning towards General Medical Image Segmentation" (2020). Electronic Thesis and Dissertation Repository. 6897.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License