Non-invasive Screening Of Esophageal Carcinoma Based On Swallowing Vibration

The incidence and mortality of esophageal carcinoma in China is high,which seriously affects people's living standards.At present,most examination methods of the esophageal carcinoma used in clinical practice are invasive,and expensive,which results in a tow acceptance of patients.These problems seriously hinder the prevention and treatment of esophageal carcinoma.A non-invasive,tow-cost detection method helps people to evaluate their esophagus at any time,and discover the lesions in time,and improve their survival rate.In this paper,a non-invasive detection system for esophageal carcinoma based on swalowing vibration signal is proposed.An accelerometer was used to coliect the swallowing vibration signal from the surface of human body,and a SVM classification model was designed to achieve non-invasive screening of esophageal carcinoma.The specific research contents and conclusions are as follows:(1)On the basis of a fiurther study of swallowing signal detection theory and esophageal physiology,a non-invasive detection system for esophageal carcinoma based on swaDowing vfibration was built.The system was conposed of a lower cornputer with a single chip as the core and a host cimputer based on the LabVIEW platform The vibration signal acquisition scheme was designed and the detection method,position,etc.,were determined.(2)A series of pre-processing operations on the raw data were performed.Firstly,a high-pass FIR filter was designed to eliminate the baseline drift caused by breathing,heartbeat and other pseudo-actions.Then,the discrete Meyer wavelet was used to reduce the high-frequency no:ise.Finally,the signal was normalized by Z-score means and was segmented into different regions according to the local standard deviation The results showed that the swalowing signals collected at the first stenosis(N1)was the strongest,while signals collected from the second(N2)and the third(N3)stenosis were greatly affected by heartbeat and respiratory,and led to relatively low s:ignal-to-noise ratio.Effective feature extraction wasdifficult to be conducted to signals collected from N2 and N3.(3)Features extraction and analysis were conducted to the pre-processed first stenosis signal,including power spectrum,band relative intensity increment,characteristic frequency values.According to the results,the relative intens it y increment of patients was different from that of the healthy group in multiple frequency bands.Furthermore,the frequency band of the patient's maximal increment was shifted to the left of the healthy group.For the peak frequency,patients showed a higher value compared to healthy grop.In addition,the signals of male patients with different disease stages were analyzed,and the results showed that the peak frequency of patients increased with the deterioration of the disease.(4)A SVM diagnostic model was built based on the extracted features.The SVM classification model was established by using the relative intensity increment of the frequency band,the characteristic frequency values and the gender as the feature vectors.The performance of the modified model was evaluated using the data from the test set.The results implied that the two-category model had an accuracy of 82O%,and the average F1-score of the model was 82%,of which the health recognition rate reached 89%.In particular,the model can distinguish male patients from health group with a sensitivity of 89%.In summary,there is a significant difference in the frequency domain of swallowing vibration signals between the patient group)and the healthy group.The preliminary analysis indicates that the difference has universal characteristics.Our research proved that the detection of swallowing vibration signal has a good application prospect as a non-invasive screening method of esophageal carcinoma.