Research On Small Object Detection In Medical Image Analysis

Small target detection in medical images is one of the hotspots in the field of medical image processing,and is of great significance for diagnosing diseases.However,due to the small size,large number,and unclear grayscale features of small targets,manual labeling is time-consuming and labor-intensive,and may introduce inconsistency.Therefore,using computer algorithms to automatically label small targets in medical images can effectively reduce the burden on doctors and improve work efficiency.Currently,existing medical image processing methods cannot accurately identify small targets,and have low accuracy in complex cases.Therefore,this thesis aims to develop new AI algorithms to realize two typical small object detection tasks in OCT imaging,and to achieve computer-aided diagnosis with an accuracy close to the level of human experts.The specific research content of this thesis is as follows:(1)Stent detection with very thick tissue coverage in intravascular OCTCoronary stenting or percutaneous coronary intervention(PCI)is widely used to treat coronary artery disease.Improper deployment of stents may lead to post-PCI complication,in-stent restenosis,stent fracture and stent thrombosis.OCT with micron-scale resolution provides accurate in vivo assessment of stent apposition/malapposition and neointima coverage.However,manual stent analysis is labor intensive and time consuming.Existing automated methods with intravascular OCT mainly focused on stent struts with thin tissue coverage.We have developed a novel deep learning method that combines UNet and convolution-based upsampling modules to automatically extract detailed features of thin neointimal coverage(≤ 0.3 mm)and very thick neointimal coverage(>0.3mm)on metal stents for stent detection,and an algorithm to accurately analyze stent area for vessels with multiple stents.25203 images from 56 OCT pullbacks and 41 patients were analyzed.Three-fold cross-validation demonstrated that the algorithm achieved a precision of 0.932±0.009 and a sensitivity of 0.939±0.007 for stents with ≤0.3 mm tissue coverage,and a precision of 0.856±0.019 and a sensitivity of 0.874±0.011 for stents with >0.3 mm tissue coverage.The correlation between the automatically computed and manually measured stent area is 0.954(p<0.0001)for vessels with a single stent,and is0.918(p<0.0001)for vessels implanted with multiple stents.The proposed method can accurately detect stent struts with very thick tissue coverage and analyze stent area in vessels implanted with multiple stents,and can effectively facilitate the evaluation of stent implantation and post-stent tissue coverage.(2)Automatic measurement of anterior chamber angle parameters in AS-OCT images using deep learningThe early assessment of angle closure is of great significance for the timely diagnosis and treatment of primary angle-closure glaucoma(PACG).Anterior segment optical coherence tomography(AS-OCT)provides a fast and non-contact way to evaluate the angle close using the iris root(IR)and scleral spur(SS)information.We aim to develop a deep learning method to automatically detect IR and SS in AS-OCT for measuring anterior chamber(AC)angle parameters including angle opening distance(AOD),trabecular iris space area(TISA),trabecular iris angle(TIA),and anterior chamber angle(ACA).3305AS-OCT images from 362 eyes and 203 patients were collected and analyzed.Inspired by the recently proposed Transformer-based architecture that learns to capture long-range dependencies by leveraging the self-attention mechanism,a hybrid convolutional neural network(CNN)and Transformer model to encode both local and global features was developed to automatically detect IR and SS in AS-OCT images.Experiments demonstrated that our algorithm achieved a significantly better performance than state-of-the-art methods for AS-OCT analysis with a precision of 0.941,a sensitivity of 0.914,a F1 score of 0.927,and a mean absolute error(MAE)of 37.1±25.3 μm for IR,and a precision of0.805,a sensitivity of 0.847,a F1 score of 0.826,and a MAE of 41.4±29.4 μm for SS,and a high agreement with expert human analysts for AC angle parameter measurement.We further demonstrated the application of the proposed method to evaluate the effect of cataract surgery with IOL implantation in a PACG patient,and to assess the outcome of ICL implantation in a patient with high myopia with a potential risk of developing PACG.The proposed method can accurately detect IR and SS in AS-OCT images and effectively facilitate the AC angle parameter measurement for pre-and post-operative management of PACG.