Session: 02-02-01: Design, Modeling and Systems

Paper Number: 112558

112558 - BI-Level 3D Reconstruction of Malignant Pleural Mesothelioma Volume From CT Images 

A malignant pleural mesothelioma (MPM) is an aggressive cancer that develops in the lining of the lungs (pleura). The disease is caused primarily by the exposure to asbestos, a fibrous mineral that was widely used in a variety of industries until the 1980s. Since then, asbestos use has been heavily regulated in many countries, however, MPM remains a major public health concern due to its latency period, which can range from 20 to 50 years, and its high mortality rate. An MPM is characterized by the development of malignant tumours in the pleural lining, which may spread to other parts of the body, such as the lungs, chest wall, diaphragm, and lymph nodes. Due to the lack of specific blood tests or imaging techniques that can confirm the diagnosis of MPM, it is difficult to make a definitive diagnosis. Chemotherapy is a common treatment for malignant pleural mesothelioma. In this process, drugs are used to either kill or slow the growth of cancer cells. A patient's chemotherapy dose is usually adjusted according to the size of the tumor and the stage of the disease when treating malignant pleural mesothelioma.

The purpose of this work is to provide doctors and health care providers with an assistive system based on the volume calculation of MPM.  Research, design, and implementation of the developed system comprise of a bi-level process as follows.

The first step is to classify CT images of lungs using machine learning (ML) and deep learning (DL) techniques in order to diagnose malignant pleural mesothelioma. The work addresses challenges associated with deep nets, including the need for large and diverse datasets, the optimization of hyperparameters, and the possibility of bias in the data. This is done by comparing two CNN architectures in terms of their features and performance, Inception-v3 and ResNet-50. Furthermore, three hyperparameters  (learning rate, batch size and epochs) were optimized for  each model, producing a wide range of possible training cases for this specific task.

In the second step, CT images are used to calculate the volume of the tumor, which can provide valuable information when determining the dose of chemotherapy to administer to the patient.  A stack of 2D images is processed in order to produce a 3D model that accurately represents the anatomical structures within the imaged region. There are several steps involved in this process, including image pre-processing, cropping volume, registering and filtering images, filling holes, segmentation, and 3D reconstruction. The volume assessment is crucial for estimating the amount of cancer cells to target and adjusting chemotherapy doses appropriately.

The results of this study demonstrate that the system developed, utilizing CNN optimizations and reconstruction of 3D images from CT images, can benefit the treatment of MPM patients.

Presenting Author: Anna Ghidotti University of Bergamo

Presenting Author Biography: She is a PhD student in Technology, Innovation and Management. Her research focuses on 3D modelling of human districts for morphological analysis, surgical planning and product customization. She is dealing with AI techniques for automated segmentation.

Authors:

Anna Ghidotti University of Bergamo
Daniele Regazzoni University of Bergamo
Miri Weiss Cohen Braude College of Engineering