Machine learning-based identification and classification of pulmonary nodules

Date of Publication


Document Type

Bachelor's Thesis

Degree Name

Bachelor of Science in Computer Engineering

Subject Categories

Computer Engineering | Electrical and Computer Engineering


Gokongwei College of Engineering


Electronics And Communications Engg

Thesis Advisor

John Anthony C. Jose

Defense Panel Chair

Armie E. Pakzad

Defense Panel Member

Carlo Noel E. Ochotorena
Lawrence Y. Materum


The current clinical practice performed in the diagnosis of lung cancer is a thorough yet lengthy procedure featuring X-ray scans, CT scans, and depending on the initial findings - lung biopsies and other kinds of imaging tests. However, these tests tend to be highly invasive and costly. An alternative to this are computer-aided diagnosis tools which involve deep learning models that fully automate the prediction in lung diagnosis, reducing or potentially eliminating the invasiveness and costliness of procedures present in the current clinical practice. This study aimed to develop a medical imaging application that allows for automatic identification and classification of the pulmonary nodules present in chest CT scans utilizing a deep learning model and utilizing the Lung Image Database Consortium Image Collection (LIDC-IDRI) dataset and its subset, LUng Nodule Analysis 2016 (LUNA16). Furthermore, the study involved the implementation of two models - Nodule Detection and Segmentation Model and Nodule Classification Model. For the segmentation model, the NoduleNet framework was used which consisted of 3 parts: candidate generation, decoupled false positive reduction, and segmentation refinement. The researchers were able to achieve a detection performance score of 89.74% and a segmentation DICE of 72.89. As for the classification model, the model architecture was based on Inception v3 and used techniques such as test-time augmentation and random oversampling to achieve an accuracy of 76.97%. Both models were successfully created and were used in the medical imaging application to assist clinicians in giving diagnoses.

Abstract Format







Lungs—Cance—Diagnostic imaging; Machine learning

Upload Full Text


Embargo Period


This document is currently not available here.