Piyush Samant

Ph.D

Mirxes Labs Pte. Ltd., Singapore

Data Scientiest

NLST (Learn more about this study)

NLST-958

Sep 8, 2022

Analyzing Pulmonary Nodules to Spot Lung Cancer in Low-Dose CT scans

Classification for lung cancer contained some ambiguity with respect to classification of lung cancer with multiple pulmonary sites of involvement. Literature says that there was marked variability in how different medical practitioners classify specific tumour, thus undermining the primary goal of classification, which is to provide a nomenclature for tumour extent that creates homogeneous cohorts of tumours. Furthermore, the heterogeneity resulting from this variability hampers the ability to interpret research studies. The symptoms of lung cancer do not appear until the disease is already at an advanced stage. Even when lung cancer does cause symptoms, many people may mistake them for other problems, such as an infection or long-term effects from smoking. This may delay the diagnosis.
In addition to this in CT Lung Cancer screening, a huge number of CT scans need to be analysed, which is enormous burden for radiologists. Therefore, from always there is a need to develop computer aided system to optimize screening and efficiency. Artificial Intelligence has proved itself in various medical imaging-based diagnosis problems as game changer. Deep Convolutional Neural Networks (CNN) have recently outperformed the previous results and performances for a variety of medical imaging tasks. In last 3-4 years several CNN techniques are proposed by researchers for the lung Nodule segmentation.
A key component of lung cancer screening is evaluating the risk of malignant lung nodule. The challenge of distinguishing malignant nodule from benign nodule has also drawn attention from many researchers. The diagnosis accuracy and other important parameters also need to be optimized. Because of the complex structure of the lung reducing the false positive rate is a crucial task. Another important aspect of the lung are blood vessels, it is difficult to distinguish between benign and malignant nodules because the density of inflammatory tissue is similar as that of tumour tissue, and both show a high level of glucose metabolism. Also, the complications like pleural adhesion and pleural effusion make it difficult to separate benign and malignant tissue.
Currently available CAD systems use segmentation techniques that eliminate the peripherally placed nodules along with the outer region of the chest. Hence, such systems detect nodules only in the interior region and tend to miss the peripherally placed nodules. Thus, efficient segmentation of peripheral nodules is a challenging task. These are the issues that motivated the research proposal. Henceforth, this research work aims at improving the performance of the CAD system for the detection of lung nodules thereby increasing the accuracy of lung cancer diagnosis. For our research, we need the clinical data, follow-up data, CT images, and pathology slides that are in NLST. The NLST data will be used to train and test the proposed methods.

The ultimate goal of the study is to develop and validate an AI-based Computer Added Diagnosis (CAD) that achieved higher performance in diagnosing the lung cancer from lung CT scans. Also, to automated quantification of non-contrast lung CT for Lung Cancer analysis. In the proposed framework the following section will be covered:

Diagnostic support: Optimize the performance parameters like sensitivity, specificity etc. of cancer diagnosis with the help of the CT imaging by using the most advanced Deep learning techniques.
Risk assessment: Quantification of lung involvement that is prompt, objective, and standardized for the purpose of supporting severity rating and triage.
Informed: Make better-informed treatment decisions through higher specificity for outcome prediction.
Efficient: Achieve prompt and consistent reporting on the patient's pulmonary condition and enhance resource allocation by doing more thorough patient triage.

The major aims of the study are as follows:
• To develop and validate a DL and Computer vision-based system for the segmentation of oblique fissures of lungs.
• To develop a pre-processing novel phase that enhances the various structures of lungs by morphological operations.
• To investigate the use of deep learning for the accurate detection of lung nodules.
• Work towards finding the nodule types by analysing the various other aspects of the nodule to be benign and malignant.
• To identify the stages of lung cancer using fully automatic and/or semi-automatic methods.
• To analysis the lung CT image by image as 2D analysis and as a full 3D analysis.
• Comparative and robustness analyses on the existing research in terms of the different image pre-processing techniques, segmentation, and classification systems.