Abstract:8034 Background: The histopathological prediction of malignant lung nodules is crucial for preoperative planning, but it always remains not precise until the detailed pathological evaluation is performed after the surgery. Thus, to define the histology types (in situ adenocarcinoma (AIS), microinvasive adenocarcinoma (MIA), and invasive adenocarcinoma (IA)) of pulmonary adenocarcinoma appearing as lung nodules before the operation and reduce unnecessary invasive diagnosis and treatment operations, we developed a classification model of based on CT images. Methods: Patients who were diagnosed with pulmonary adenocarcinoma (tumor diameter ≤ 3cm) at Nanfang Hospital, Southern Medical University, China, were retrospectively enrolled, including detailed pathology results, computed tomography (CT) images, and basic information. Through training the model proposed in our previous study, we proposed a new model for defining histopathological types. Then, we used this data to train and compare the diagnostic efficacy of our model with another previously reported deep learning (DL) model (that was a segmentation - classification model with the 3D Unet++ - ResNet-50 combined model) and machine learning (ML) model (that was a traditional model Grey Level Co-occurrence Matrices (GLCM) as the feature extractor and the support vector machine (SVM) as the classifier) in the same method. Results: A total of 2061 patients (with 395 of AIS, 334 of MIA and 1332 of IA) were retrospectively enrolled from January 2019 to April 2022. The data from 2/3 of the patients were randomly selected for training and another 1/3 for verifying. Through training and verifying, our model shows much better diagnostic efficacy, with areas under receiver operating characteristic (ROC) curves with 95% confidence interval (CI) of 0.97(0.96, 0.98), comparing with those of the ML model of 0.77(0.73, 0.81) and the DL model of (0.90(0.88, 0.92). The heatmap visualization of the proposed model shows the classification ability with prediction probabilities of our model clearly and understandably. Conclusions: In this study, we developed a novel artificial intelligence multi-task learning model to precisely predict the pathology type of malignant lung nodules with satisfying diagnostic efficacy, which might provide a new and reliable way for preoperative planning of lung nodules. Further external verifying should be done in the future. In addition, prediction of the growth patterns of pulmonary adenocarcinoma could be more helpful in addition to predicting histopathological types.

Benign and Malignant Classification Model of Pulmonary Nodules Based on Residual Neural Network

Classification of Lung Nodules Based on Deep Residual Networks and Migration Learning

Multiclassifier Fusion Based on Radiomics Features for the Prediction of Benign and Malignant Primary Pulmonary Solid Nodules

A Simple Method to Train the AI Diagnosis Model of Pulmonary Nodules.

Deep-Learning Model of ResNet Combined with CBAM for Malignant–Benign Pulmonary Nodules Classification on Computed Tomography Images

Explainable Classification of Benign-Malignant Pulmonary Nodules With Neural Networks and Information Bottleneck

Application of Computer Deep Learning Model in Diagnosis of Pulmonary Nodules

Predictive Model for the Diagnosis of Benign/malignant Small Pulmonary Nodules

Classification of Benign and Malignant Lung Nodules Based on Deep Convolutional Network Feature Extraction

A diagnostic classification of lung nodules using multiple-scale residual network

Interpretative Computer-aided Lung Cancer Diagnosis: from Radiology Analysis to Malignancy Evaluation

Artificial intelligence-based prediction model of malignant lung nodules for preoperative planning.

Models of Artificial Intelligence-Assisted Diagnosis of Lung Cancer Pathology Based on Deep Learning Algorithms

Predicting Benign, Preinvasive, and Invasive Lung Nodules on Computed Tomography Scans Using Machine Learning

Pulmonary nodule classification with deep residual networks

Machine Learning Model of ResNet50-Ensemble Voting for Malignant–Benign Small Pulmonary Nodule Classification on Computed Tomography Images

Evaluate the Malignancy of Pulmonary Nodules Using the 3D Deep Leaky Noisy-or Network

Diagnostic performance of a deep learning-based method in differentiating malignant from benign subcentimeter (≤10 mm) solid pulmonary nodules

Towards Reliable and Explainable AI Model for Solid Pulmonary Nodule Diagnosis

Deep learning predicts malignancy and metastasis of solid pulmonary nodules from CT scans

A Self-supervised Learning-Based Fine-Grained Classification Model for Distinguishing Malignant From Benign Subcentimeter Solid Pulmonary Nodules