Multi-Site Automated Segmentation and Multi-Parametric MRI Quantification to Assess the Effect of Treatment of Venous Malformations
Abstract
Topic Area: Vascular Malformations The human body is composed of a complex network of blood vessels that are crucial for circulating oxygen and nutrients to organs. Vascular malformations are a type of abnormally formed blood vessels that usually develop before birth and persist throughout life. The most common subtype is venous malformations (VMs), which specifically involve veins and lymph vessels, and affect about 1 in 100 people. Most venous malformations happen sporadically and without any underlying cause. They can involve any body part and affect patients of any age as they grow throughout life. As a result, patients can suffer from various symptoms, including debilitating pain, functional loss (e.g., limping, limited movement), cosmetic deformity (and social stigma), swelling, bleeding, or weakness. VMs are currently treated with a local image-guided treatment called sclerotherapy, which includes injection of certain agents into the lesion (abnormal region) that causes local inflammation and eventual reduction of the volume of the lesion. In recent years, there have been advances in both image-guided and pharmaceutical treatments for VMs with the introduction of new systemic therapies. However, the effects of these therapies, either standard or new, are incompletely understood because there are no objective ways to assess treatment outcomes. Most clinical assessment is highly subjective, with physicians interpreting patients’ descriptions of ongoing symptoms or changes in those symptoms after treatment. Recently, the first clinical outcome measurement tool has been published for the assessment of treatment outcomes in patients with VMs. Using this tool, patients report the changes in their quality of life without the influence or interpretation of a clinician. VMs differ widely in location, size, extent, and imaging characteristics across patients. Therefore, an objective assessment of treatment outcome is considered challenging especially if a lesion was treated in multiple sessions and/or with different agents across time. We believe that magnetic resonance imaging (MRI) will provide a robust and objective outcome measure for VMs. However, MRI images consist of many different adjustable components, and it is not established which changes in these components may be best used to assess VMs. Further complicating this, the variability, both within and between VMs, makes it more challenging to decide which properties are most relevant to the disease burden and what changes represent a beneficial response to treatment. Therefore, an understanding of how VMs change in a way that is detectable under MRI will allow physicians to follow lesions over multiple sessions of therapy without irradiating the patient (who are often children). Furthermore, the use of automatic analysis tools, developed through artificial intelligence, should allow us to rapidly and uniformly assess complex VMs. We hypothesize that an automated imaging method of analyzing magnetic resonance images can be used to objectively assess the treatment outcomes for VMs. Therefore, our experimental aims are as follows: Aim 1: Create and validate a neural network (NN) to segment VMs on MRI with varying acquisition parameters, based on a combined GAN synthesis NN for image signal normalization and a U-Net NN for segmentation. Aim 2: To quantify MRI radiomics in the segmented regions of the VMs over time, across treatments, and at two independent sites on T2-weighted and apparent diffusion coefficient (ADC) images. Aim 3: To quantitatively correlate the MRI findings and patient-reported clinical outcomes. This proposal is innovative in (1) its goal to develop the first-ever fully automated assessment tool for VMs; (2) providing a quantitative assessment of MRI changes occurring in VMs over serial treatments; and (3) determining the correlation between MRI findings and clinical outcomes. As there has been no standardized method for assess
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2024
- Source ID
- HT94252310032
Entities
People
- Craig Jones
Organizations
- Johns Hopkins University
- United States Army