Classification of Benign and Malignant Lung Nodules Based on CT Raw Data

Image, Body, Lung Cancer

radiomics, lung cancer, classification, CT, raw data

The employ of medical images combined with deep neural networks to assist in clinical diagnosis, therapeutic effect, and prognosis prediction is nowadays a hotspot. However, all the existing methods are designed based on the reconstructed medical images rather than the lossless raw data. Considering that medical images are intended for human eyes rather than the AI, we try to use raw data to predict the malignancy of pulmonary nodules and compared the predictive performance with CT. Experiments will prove the feasibility of diagnosis by CT raw data. We believe that the proposed method is promising to change the current medical diagnosis pipeline since it has the potential to free the radiologists.

The routinely used diagnostic scheme of cancers follows the process of signal-to-image-to-diagnosis. It is essential to reconstruct the visible images from the signal of medical device so that the human doctor can perform diagnosis. However, the huge amount of information inside the signal is not optimally mined, which causes the current unsatisfactory performance of image based diagnosis. In this clinical trial, we will develop an AI based diagnostic scheme for lung nodules directly from the signal (raw data) to diagnosis, skipping the reconstruction step. In this trial, we will focus on the discrimination of malignant from benign lung nodules. We will collect a dataset of patients who are screened out lung nodules. All patients undergo preoperative CT scan (raw data and CT images available) and have pathologically confirmed result of the nodules. We will build a model using only raw data for diagnosis of the lung nodules. Moreover, another model from CT image will be built for comparison. Furthermore, we will perform follow-up on these patients and build a model based on CT raw data for prognosis analysis of lung cancer.

China