Knowledge-Based Analysis for Mortality Prediction From CT Images

Abstract

Low-Dose CT (LDCT) can significantly improve the accuracy of lung cancer diagnosis and thus reduce cancer deaths compared to chest X-ray. The lung cancer risk population is also at high risk of other deadly diseases, for instance, cardiovascular diseases. Therefore, predicting the all-cause mortality risks of this population is of great importance. This paper introduces a knowledge-based analytical method using deep convolutional neural network (CNN) for all-cause mortality prediction. The underlying approach combines structural image features extracted from CNNs, based on LDCT volume at different scales, and clinical knowledge obtained from quantitative measurements, to predict the mortality risk of lung cancer screening subjects. The proposed method is referred as Knowledge-based Analysis of Mortality Prediction Network (KAMP-Net). It constitutes a collaborative framework that utilizes both imaging features and anatomical information, instead of completely relying on automatic feature extraction. Our work demonstrates the feasibility of incorporating quantitative clinical measurements to assist CNNs in all-cause mortality prediction from chest LDCT images. The results of this study confirm that radiologist defined features can complement CNNs in performance improvement. The experiments demonstrate that KAMP-Net can achieve a superior performance when compared to other methods. Our code is available at https://github.com/DIAL-RPI/KAMP-Net.

Deep learning predicts cardiovascular disease risks from lung cancer screening low dose computed tomography

“…With a similar sensitivity of 96.6%, our model achieved a slightly but not significantly higher PPV of 38.4% ( p = 0.3847). In addition, we compared our model with the three recently reported works, KAMP-Net 33 , Auto-encoder (AE+SVM) 32 , and a deep learning based CAC socring model (DeepCAC) 17 . The table in Fig.…”

“…To evaluate the generalization ability of the deep learning quantified CVD risk score, we directly applied the trained model to the MGH data and evaluated the consistency between the model-predicted risk score from LDCT and the three clinically adopted risk scores calculated from ECG-gated cardiac CT. Our model was compared with two other previously reported studies on CVD risk prediction 32 , 33 .…”

“…First, while a chest LDCT image volume has a large field view, the heart takes only a small subset. Features extracted by a deep learning model on the entire image may hide the CVD risk related information 33 . To tackle this problem, a cardiac region extractor was developed to locate and isolate the heart so that the risk estimation model can focus on this region of interest.…”

“…However, such a two-stage method may not be able to extract distinctive features associated with CVD, since the first stage has little knowledge about the final objective. Our previous work showed the feasibility of predicting all-cause mortality risk from patient’s LDCT images from 180 subjects 33 . The developed method, KAMP-Net, first selects a representative 2D key slice from the whole LDCT volume, and then applies an end-to-end CNN to predict all-cause mortality risk (area under the curve (AUC) of 0.76).…”

Artificial intelligence for detection and characterization of pulmonary nodules in lung cancer CT screening: ready for practice?

“…At present, there are no AI algorithms that focus on this by for example predicting the histological subtype of screening-detected pulmonary nodules, the growth rate, or the metastatic potential of pulmonary nodules. Though there are algorithms which were designed to predict the time of death from a scan (66)(67)(68)(69), there is a lack of appropriate data to perform studies which attempt to predict the risk that lung cancer progression will be the cause of death.…”