LMLCC-Net: A Semi-Supervised Deep Learning Model for Lung Nodule Malignancy Prediction from CT Scans using a Novel Hounsfield Unit-Based Intensity Filtering
Tasnia Binte Mamun, Adhora Madhuri, Nusaiba Sobir, Taufiq Hasan
TL;DR
This work tackles the automated malignancy prediction of lung nodules in CT by exploiting Hounsfield Unit (HU) distributions through a novel Learnable HU-based intensity filtering scheme implemented as a multi-branch 3D CNN (LMLCC-Net). It introduces a semi-supervised labeling approach to resolve ambiguities in the LUNA16 dataset and a Learnable Dynamic Range Layer that optimizes HU ranges during training, yielding improved discrimination between benign and malignant nodules. Empirical results on LUNA16 show strong performance with $A_{CC}$ around $91.96\%$, $S_{EN}$ about $92.94\%$, and $AUC$ near $94\%$, outperforming several prior methods and demonstrating clinical utility for radiologists. The combination of robust labeling, HU-focused feature extraction, and interpretable Grad-CAM analysis supports potential deployment in real-world screening and diagnostic workflows.
Abstract
Lung cancer is the leading cause of patient mortality in the world. Early diagnosis of malignant pulmonary nodules in CT images can have a significant impact on reducing disease mortality and morbidity. In this work, we propose LMLCC-Net, a novel deep learning framework for classifying nodules from CT scan images using a 3D CNN, considering Hounsfield Unit (HU)-based intensity filtering. Benign and malignant nodules have significant differences in their intensity profile of HU, which was not exploited in the literature. Our method considers the intensity pattern as well as the texture for the prediction of malignancies. LMLCC-Net extracts features from multiple branches that each use a separate learnable HU-based intensity filtering stage. Various combinations of branches and learnable ranges of filters were explored to finally produce the best-performing model. In addition, we propose a semi-supervised learning scheme for labeling ambiguous cases and also developed a lightweight model to classify the nodules. The experimental evaluations are carried out on the LUNA16 dataset. The proposed LMLCC-Net was evaluated using the LUNA16 dataset. Our proposed method achieves a classification accuracy of 91.96%, a sensitivity of 92.94%, and an area under the curve of 94.07%, showing improved performance compared to existing methods The proposed method can have a significant impact in helping radiologists in the classification of pulmonary nodules and improving patient care.