Chromosomes are intracellular aggregates carrying genetic information in genes, which are major objects of study in biological cytogenetics. An abnormal number or structure of chromosome causes chromosome disorders. Chromosome screening is an important part of prenatal the care. However, manual analysis of chromosomes is time-consuming. With the increasing popularity of prenatal diagnosis, manpower becomes overstretched. Therefore, an automatic approach for chromosome detection and recognition is urgently needed. Here, we proposed a vision transformer-based method for automated chromosomal object detection on metaphase cell images. The proposed method, named Swin Transformer Chromosomes Net (Swin-TCNet), that was mainly composed of Swin Transformer, region proposal network (RPN) and, multi-intersection over union (IoU) strategy. A large database formed by 5,000 metaphase cell images consisting of a total of 229,852 chromosomes was used to develop and evaluate the proposed method. The method uses the original cell meta images without pre-processing as input because that is clinically more friendly. The evaluation method used is the COCO API's mAP50, the testing set is up to 1250 original cell meta images (57,484 chromosomes in total), which is much more than the existing related literature. The accuracy and F1-score of our proposed method are 98.9% and 99.1%, respectively. Moreover, the practicality and robustness of Swin-TCNet in clinical practice can also be confirmed through a large amount of validation data.

Sleep apnea is the most prevalent sleep disorder. In severe cases, it can even lead to sudden death. To diagnose sleep apnea, it is critical to measure the number of sleep arousal times per hour. Current clinical standard procedures require patients to visit a sleep center or hospital for polysomnography (PSG), in which the patient's physiological signals are recorded during the night, and a sleep technician then interprets these signals manually to determine the sleep-wake pattern. With the rising number of sleep disorders, sleep technologists cannot afford such a high workload, which makes the development of an automated sleep arousal detection system critical to reducing medical costs. When PSG is recorded, a large number of wires will also interfere with the patient and cause measurement distortion. In this study, a spatial-channel attention (SCA) U-Net for automatic sleep arousal detection was proposed, which was trained and tested on PhysioNet 2018 sleep PSG dataset. Considering clinical applicability and reducing patient disturbance when recording physiological signals, only five physiological signal channels were required for our proposed model. It is much lower than other methods in the literature. The A AUPRC of our proposed model was and 39%. The results showed the proposed method is highly accurate and requires fewer channel signals, that also demonstrated the clinical applicability and robustness of the proposed method.