| Spectral Domain Optical Coherence Tomography（SDOCT）is an optical tomography imaging technology with high resolution and large depth.SDOCT is based on optical coherence tomography（OCT）.Compared with Time Domain Optical Coherence Tomography（TDOCT）,SDOCT has higher imaging speed and sensitivity,so it can be widely used in the field of ophthalmology imaging,and at the same time has a wide application prospect in the direction of human tissue imaging in other medical fields.However,the axial resolution decreases with the increase of imaging depth,and the distortion caused by unstable optical path structure still limits the application of SDOCT technology in the field of ophthalmic imaging.In addition,there is currently no intelligent diagnosis algorithm that can realize auxiliary diagnosis and treatment,which greatly reduces the efficiency of diagnosis,and may cause missed diagnosis and misdiagnosis caused by subjective reasons.In view of a series of problems existing in SDOCT equipment,this paper focuses on the key technologies of intelligent SDOCT system,and conducts research on system development and intelligent diagnosis algorithm.The main research contents include the following three items:1.In this paper,we use Zemax optical simulation software to design and simulate the core optical part of the intelligent SDOCT system,mainly including the structure of the sample arm and the spectrometer.In the sample arm structure,a two-dimensional scanning galvanometer is used to scan the sample,and a scanning range of 8mm×8mm can be achieved by adjusting the scanning angle.In the structure of the spectrometer,the grating is used to split the light,and the problems caused by chromatic aberration are eliminated through four achromatic lenses,and the CCD camera is used for imaging.The simulation results verify the rationality of the optical path structure of the system.2.In this paper,a small and intelligent SDOCT principle prototype is designed and developed,which mainly includes the structure of interferometer and spectrometer.The imaging depth is 2.7mm,and the axial resolution and lateral resolution are 5μm and 14.5μm respectively,which meets the needs of ophthalmic imaging.The scan rate reaches 12.5k Hz,and the scan range is 8mm×8mm.In the light source module,in order to ensure the stable operation of the laser module for a long time,a temperature control circuit of the laser is developed,and the PID algorithm is used to control the temperature of the entire light source,so that the laser can maintain the working temperature of 25°C for a long time,and the accuracy is±0.1°C.We collected onion tissue OCT images and human hand paronychia tissue OCT images respectively,and achieved good imaging capabilities.3.This paper uses the Wide-ECANet deep learning network model to design and manufacture a set of intelligent diagnosis algorithms for ophthalmic diseases,which can diagnose and screen ophthalmic diseases such as AMD,DME,DRUSEN,and CNV,and optimize the models from different perspectives such as accuracy,diagnosis speed,and sensitivity.Two authoritative ophthalmic OCT image datasets with different amounts of data are used to train the model,and the current advanced deep learning network（SOTA）,other channel attention mechanism neural network and the diagnosis results of authoritative physicians are used as Contrast experiments to verify the advanced degree of the network.The experimental results show that the diagnostic accuracy rates of the Wide-ECANet network on the two datasets are 1.0 and 0.985,respectively,which are of high level.Compared with other deep learning networks,the performance of this network is better,it can reach the diagnostic level of authoritative doctors,and it has a faster diagnosis speed.In addition,this network can effectively alleviate the confusion problem that occurs when different diseases are diagnosed. |
PDF Full Text Request