If the TIS system and manual screening are used simultaneously in cytological interpretation, the computing ability of the computer and the interpretation of the human brain are both used [21], that means combining computer-aided preliminary screening and human experience.The advantages of this practice is to make full use of computers in identifying DNA density [22], computer can automatically screen parts of a normal smear and avoid manual screening, thus reducing the labor of the human brain [23].
Retrospective and prospective studies have shown that TIS systems can significantly improve efficiency[24]. TIS system can improve the efficiency by 38.9% ---252% and significantly shorten the screening time comparing to the traditional way [25]. Comparing with manual screening, the TIS system effectively reduces the area that cytological doctors screen, therefore it concentrates more on the accurate interpretation level of cytopathic cells [26].
The imaging instrument and review scopes of the TIS system can effectively improve the working efficiency of TCT laboratories and their ability to identify abnormalities [27]. Therefore, TIS can maximize the efficiency and accuracy of the laboratory in cervical cancer detection, which increases the sensitivity, and has a higher PPV than manual screening for ASCUS and above lesions [13]. With a larger sample size of 257,047 cytological screenings and 56,484 pathological biopsy data, our data were basically consistent with those of literature reports.
We analyzed the age distribution of TIS and manual groups. There was no statistically significant difference in age distribution between the two groups ( Table 1, t-test, t =0.99, P > 0.05), indicating that age was not a confounding factor between two groups.
For LSIL lesions, comparing with manual screening, the positive detection rate of TIS was increased by 36.5%, with statistically significant differences (P < 0.0001), which is consistent with other literature. For ASCUS, AGC-FN and HSIL lesions, the positive detection rates of the TIS group decreased significantly (P < 0.0001). In TIS group, ASCUS cases were accounting for 11.83% of the total cases, significantly lower than the manual group 12.97 %. However, the positive rate of ASCUS biopsy in the TIS group (27.20%) was significantly higher than the manual group (21.06%) (P < 0.0001), therefore, we consider that the TIS group can reduced the uncertain TCT results and classified them into LSIL, so that the detection rate of ASCUS lesions was reduced, but the specificity of TIS group was higher than manual group. Due to the low incidence rate of SCC, AIS and IAC and their typical cytological morphology, SCC, AIS and IAC were more easily interpreted, so the detection rates of SCC, AIS and IAC did not change significantly between the TIS group and the manual group (Table 2). Meanwhile, few literature have discussed whether the detection rate of TIS for SCC, AIS and IAC is improved, so we can increase our sample size and conduct in-depth research in the future. ASC / SIL is a interpretation ability evaluation method for TCT, and monitoring the change in its ratio is a simple and effective tool for quality control, satisfactory ASC / SIL ≤ 3 [28]. Especially in the case of new screening methods and new employees, such as the emergence of new staining methods, the ratio of ASC/SIL may increase owing to the inadaptability of interpretation[29], therefore it is necessary to train doctors appropriately to adapt to the new screening methods. According to Table 2, ASC/SIL were 2.85 and 2.04 respectively before and after TIS administration, and the overall positive rate increased, indicating that TIS did not increase the uncertain interpretation. The PPV indicates the proportion of TCT-positive patients with real disease, and the NPV indicates the proportion of TCT negative patients who are healthy. The PPV and NPV reflect the benefits of TCT for cervical cancer screening. As shown in Table 5, NPV in the manual group was 83.38%, and the TIS group was 85.20%, although there was no significant difference between the two groups, it was still increased in the TIS group. During January 2019 and December 2021, our hospital was under renovation, the number of follow-up and biopsy patients were decreased, therefore, in the TIS group, the proportion of biopsy of positive patients decreased during these period (Table 3 -4). Even though, compared with the manual group, the TPV for LSIL on histology and PPV, APV and TPV for HGD in the TIS group were improved, and the difference was statistically significant. TIS can identify patients with cervical lesions more accuracy, and reduce the disease burden of patients[30], comparison of the PPV, APV and TPV between the two groups may answer the question if TIS improved the screening efficacy.
The FNR also known as the missed diagnosis rate or class II error, refers to the percentage of patients who are actually ill but are determined to be disease-free according to the screening test. The FNR reflects the situation of patients missed in the screening test. The reasons for the FNR of TCT may be as follows: first, cervical sampling error caused by doctors. Summarizing the sampling experience and improving the sampling skills will help to reduce the incidence of false negative cases; second, patient's age, lesion location and lesion degree can lead to the occurrence of cytological false negative interpretation; third, fixation dyeing is not standardized. Concurrently, we observed that in cases of cytological abnormalities, the cell types were mainly middle surface squamous cells and outer bottom squamous cells. Most abnormal cells were scattered. The number of abnormal cells in most of the smears was less than 10. Simultaneously, nuclear enlargement and light staining were common, and the background was mostly accompanied by inflammatory changes or keratotic changes. Cytologists should more carefully identify some minimal abnormal morphological changes in smears and continuously accumulate film reading experience, which will help reduce the occurrence of false negative interpretation. In cervical cancer screening, we hope that patients with cervical lesions can be clearly screened by reducing the false negative rate. Compared with manual screening, TIS apparently reduced the false negative rate from 35.53% to 19.71% (table 5, χ2 =248.77, P < 0.0001), so as to improve the accuracy of TCT in cervical cancer screening. The FPR
was increased significantly in the TIS group, which is related to the loss of TCT positive patients and the decrease of biopsy rate in this group. With the advent of computer science and the era of data collection, artificial intelligence (AI) has gradually begun to extend and intermingle into various fields of medicine, especially in the field of cellularinterpretation. First, AI can complete cervical cell collection, fixation, and DNA staining, reducing random errors caused by manual operation. Second, similar to the TIS system in our current study, AI applies a deep segmentation network to extract regions of interest in cytopathological images and then combines the physician's experience to train a deep classification network with analysis capability. Its classification results are used to construct features, and a decision model is used to complete the grading of cytopathological images. Finally, the automatic screening of cervical fluid-based cytology images can be completed. Compared with the TIS system, artificial intelligence can gradually improve the computer algorithm model while continuously simulating the deep learning of human experience, so that the cytodiagnosis effect is also continuously improved. Therefore, we highly recommend combining the current TIS system with artificial intelligence to improve cytodiagnostics better.
In general, TIS which combines the ability of the human mind and artificial intelligence, creates an excellent cytological screening method, which significantly reduces the screening time for doctors, and selects individual small diseased cells, thus increasing the doctors' confidence in cytological screening [31]. Quantitative analysis of pathological images using computers and intelligent interpretation is an important direction for the quantitative interpretation of pathology. Cytological screening systems similar to TIS have become increasingly popular, and the successful application of TIS automatic screening in TCT will provide new ideas and methods for the development and application of intelligent interpretation of pathology.