iDeepWise.AI 'Qiaosi' by DeepThinking AI Demonstrates Groundbreaking Cervical Cancer Screening Capability on CCTV-1's 'Super Brain'

On average, one woman dies from cervical cancer every two minutes worldwide. The most effective strategy to combat cervical cancer, the second leading cause of cancer-related deaths among women, is to intervene before precancerous lesions progress to invasive cancer. With timely early screening, the cure rate for cervical cancer can reach as high as 90%. However, the survival rate for patients with advanced-stage cervical cancer is only 3%.

Regular early screening can be said to be the dividing line between life and death for cervical cancer patients.

However, in China, the cervical cancer screening rate among women remains far below the 80% coverage target stipulated in the Outline for the Development of Chinese Women (2011–2020).

For instance, in grassroots regions such as Shangrao, Jiangxi Province, China, approximately 2.2 million of the 3.31 million permanent female residents require cervical cancer screening, yet there are only five physicians in the entire city with the capacity to perform such screenings.He Weihua, a pathologist at a primary care hospital in Shangrao, Jiangxi Province, lamented, “If doctors were to perform all early screening tasks alone, it would take approximately 70 years to complete the screening.”

It is evident that the primary reason lies in the shortage of pathologists, as training a new pathologist typically takes more than eight years.

Artificial intelligence may help humanity tip the scales between life and death once again by improving early screening and diagnosis rates for cervical cancer.

In the latest episode of CCTV-1’s “Super Brain,” ten pathology experts from renowned Chinese tertiary hospitals—including Peking Union Medical College Hospital, Beijing Friendship Hospital, Women’s Hospital School of Medicine Zhejiang University, and Peking University People’s Hospital—were assembled on-site to jointly evaluate Qiaosi, an AI-powered robotic assistant for cervical cancer screening.

In the program, Dr. He Weihua, assisted by iDeepWise.AI’s Qiaosi Robot, competed alongside 10 senior pathologists in evaluating 30 cervical cytology samples to assess the screening capabilities of artificial intelligence.

On-site results indicate that, through the iDeepWise.AI-Qiaosi Robot,Primary care physicians can achieve a “one-to-ten” efficiency, delivering diagnostic results fully consistent with those of senior pathologists within the same timeframe.Among them, the iDeepWise.AI-Qiaosi Robot is from a company named “iDeepWise Artificial Intelligence.”

The theme of this episode of "Amazing Intelligence" is “Paying Tribute to Pathologists with ‘Intelligence’.” The program also invited Professor Bu Hong, a professor in the Department of Pathology at West China Hospital of Sichuan University, Executive Director of the Chinese Anti-Cancer Association, and Chairman of the Chinese Society of Pathology, to serve as the “Tribute Witness.”

As one of the first pathologists after the reform and opening-up, Professor Bu Hong has witnessed the critical role that improved quality of cytopathology screening plays in early cancer detection.

Professor Bu Hong stated, “The Department of Pathology serves all clinical departments in the hospital and is hailed as ‘the doctor’s doctor.’ Pathology is closely linked to tumor prevention. For a long time in the past, we did not have regular early cancer screening; screening was only conducted when tumors were suspected. Advances in medical technology now enable early detection of cancer, but this requires pathologists to spend considerable time observing and differentiating under the microscope.”

Cervical cancer is classified into cervical squamous cell carcinoma and adenocarcinoma. During traditional cytological screening for cervical cancer, physicians examine cell smears under a microscope and must continuously switch among three objective lenses of different magnifications. They initially scan the entire slide at 100× magnification; upon detecting abnormal cells, they switch to higher magnifications, such as 200× or even 400×, for detailed observation.

On a glass slide less than one inch in size, there are tens of thousands of cervical cells. Physicians must first screen through this large number of mostly normal cells to identify abnormal ones, and then make a diagnosis based on cellular morphology and patterns.

Among tens of thousands of cells, identifying pathological cells is akin to finding a needle in a haystack for pathologists, whereas AI systems effectively deliver the “needle” directly to them.

The iDeepWise.AI system can exclude negative cells one by one and display suspicious cells for physicians’ review, enabling them to easily determine whether the cells are pathological and significantly reducing screening time.

Enable physicians to return to the core task of diagnosis, rather than spending excessive time identifying and ruling out negative cells, thereby alleviating the current shortage of pathologists. In China, 100 million women require cervical cancer screening, yet there are only approximately 10,000 licensed pathologists nationwide.

For individual pathologists, artificial intelligence reduces workload. For the pathologist community as a whole, AI captures and consolidates clinical expertise, thereby reducing missed diagnoses and misdiagnoses.

Professor Bu Hong stated, “Pathological diagnosis relies heavily on the accumulation of physicians’ experience; however, pathologists face an intense workload. There is a saying in our industry: ‘By the time a pathologist has gained sufficient experience, they are likely to have developed cataracts.’ In contrast, artificial intelligence can continuously learn and accumulate knowledge, serving as an ‘immortal physician.’”

The initial intelligence of artificial intelligence also comes from doctors. The iDeepWise.AI Qiaosi system’s learning foundation is derived from iDeepWise’s database of 1.2 million cervical cell samples, with the data center adding nearly 1,000 new samples daily.

In addition to training on millions of high-quality sample datasets, iDeepWise has implemented substantial core-level algorithmic innovations in AI. By adapting to various slide preparation methods and scanning devices, and by integrating artificial intelligence technologies such as deep learning, image instance segmentation, transfer learning, and visual-semantic understanding, it achieves fully automated auxiliary screening for cervical cytology. The system can interpret liquid-based cytology slides prepared using TCT (membrane filtration), LCT (sedimentation), and TLT (centrifugation) methods, providing reliable test results.

Taking the globally recognized and most authoritative Herlev dataset for cervical cell images as an example, the previous best performance was reported in a recent paper by the U.S. National Institutes of Health (NIH). However, the iDeepWise AI team, leveraging innovative AI algorithms, has comprehensively surpassed these results on the same dataset under identical evaluation criteria. Notably, their model achieved a cell classification accuracy of 99.3% on key metrics, setting a new world-best benchmark for this dataset.

Validated through extensive clinical samples, the iDeepWise.AI CIAS system significantly improves the detection sensitivity for ASCUS and higher-grade lesions, markedly reduces the false-negative rate, and greatly minimizes missed diagnoses. Specifically, it achieves a sensitivity of 99.4% and a specificity of 98.9% for squamous epithelial cell lesions; for glandular cell abnormalities, the sensitivity is approximately 97.9% and the specificity is approximately 93.4%. The negative exclusion rate reaches 81.19%, with the false-negative rate as low as 0.006%.

It is understood that iDeepWise's AI-assisted cervical cancer screening system is the only AI product in China capable of effectively identifying glandular cell abnormalities.

On the set of *Superhuman Intelligence*, Dr. Meng Zhilan from the Department of Pathology at Peking Union Medical College Hospital stated, “I believe the most challenging aspect is dealing with overlapping cells, specifically cell clusters. We must switch the microscope to high magnification and continuously adjust the fine focus knob to visualize each cell as clearly as possible. If a definitive judgment still cannot be made, the case should be classified as a suspicious lesion, warranting further diagnostic biopsy.”

To address these "challenging cases" in image interpretation, iDeepWise.AI has pioneered a deep learning segmentation algorithm for cell clusters.

Dr. Yang Zhiming, founder of iDeepWise, stated, “Our AI-based cluster cell segmentation algorithm can separate cells from clusters in images before interpretation.”

Significant variations in stained cell images across different hospitals and institutions substantially reduce the efficiency of pathologists in slide review. iDeepWise’s pioneering stain intensity adaptive algorithm applies deep learning to stain normalization in digital pathology images, enabling automatic adjustment of stain intensity for atypically stained cell images.

In addition to this series of pioneering algorithms, iDeepWise has achieved multiple breakthroughs in the core AI healthcare technologies.

It is worth noting that among the 30 so-called cervical cancer cell images presented on the set of *Super Brain*, one was actually an image of urinary sediment. In identifying this image, Dr. Yu Xiaomeng from the Department of Pathology at Beijing Friendship Hospital, with 30 years of professional experience, detected this discrepancy. iDeepWise.AI performed equally well, rapidly identifying the abnormal cells.

Empowering cervical cancer screening at the grassroots level, where medical infrastructure is weak, through AI is also a key initiative strongly promoted by the National Health Commission.

In the Healthy China Action (2019–2030), the National Health Commission set targets in its maternal and child health indicators to increase the coverage of cervical cancer and breast cancer screening among eligible rural women to at least 80% by 2020 and at least 90% by 2030.

The National Health Commission has also clearly stated that it will use AI and other means to increase the screening coverage rate for “two cancers” (breast cancer and cervical cancer) by 27% within three years.

While iDeepWise is increasing its investment to overcome technical bottlenecks and elevate the sophistication of its AI-based cervical cancer screening technology, it is also striving to accelerate the practical deployment and application of its products.

Powered by iDeepWise’s independently developed world-first medical-grade AI chip and its AI algorithm model for pathological cytology, both microscopes and pathology scanners are equipped with “AI brains” operating at the edge and in the cloud. This system can concurrently process 480 smears, interpreting 70,000 cells per smear within 60 seconds. Its deployment is well-suited to China’s national conditions, extending affordable benefits to a wide range of primary-care hospitals and addressing the challenge of uneven distribution of medical resources.

It is reported that iDeepWise’s AI-powered cervical cancer screening system has been deployed in hundreds of hospitals, including more than 30 top-tier Grade A tertiary hospitals such as Fudan University Shanghai Cancer Center and The First Affiliated Hospital, Zhejiang University School of Medicine, providing AI-based early cervical cancer screening services to tens of thousands of users.

Back on the set of “Super Intelligence,” Dr. Yang Zhiming, the program’s honoree for AI-driven tribute to humanity, stated firmly, “Our vision is to tip the scales between life and death with artificial intelligence technology, paying ‘intelligent’ tribute to life.”

Professor Bu Hong, an on-site witness, stated, “Our assessment of whether a technology is ready for clinical deployment typically involves two criteria: first, evaluating its technical capabilities; and second, obtaining approval certification from the National Testing Center. I believe that the Qiaosi Robot by iDeepWise.AI has met the requirements for clinical deployment.”