iDeepWise Unveils World’s First Medical Imaging Dedicated AI Chip M-DPU, Powering Cervical Cancer Screening Solutions

VCBeat has learned that Huawei recently ranked 12th in a report by the renowned global research firm Compass Intelligence, making it the top-ranked Chinese AI chip manufacturer on the list. Although Chinese companies lag behind their foreign counterparts in overall chip capabilities, iDeepWise has already secured a key entry into smart healthcare with its independently developed world’s first AI chip dedicated to medical imaging, quickly sparking widespread attention and discussion within the industry.

Image source: iDeepWise

The smart healthcare sector is currently experiencing robust growth. The integration of artificial intelligence (AI) with healthcare has yielded a range of solutions for medical imaging. However, AI solutions offered by smart healthcare vendors generally face significant challenges, including difficulties in practical implementation and channel distribution, as well as the high cost of general-purpose GPU processors. Meanwhile, although the medical device market represents a trillion-yuan opportunity and manufacturers are eager to enhance their AI capabilities to boost competitiveness, the limitations of GPU processors—such as their large form factor, difficulty in integration, and lack of specialized optimization for medical AI—have become major obstacles to the AI upgrading of medical devices.

iDeepWise’s IDeepWise M-DPU, a medical-grade AI processor developed by iDeepWise Artificial Intelligence, is designed to comprehensively address the aforementioned challenges. By establishing a medical AI solution platform based on the M-DPU, it bridges the gap between these domains and embeds an “AI brain” into medical devices.

Image source: iDeepWise

In 2011, China promulgated the Outline for the Development of Chinese Women (2011–2020), which specifically emphasized the need to strengthen screening for the “two cancers” in women (breast cancer and cervical cancer). However, the current national situation is characterized by a shortage of radiologists, high workload in image interpretation, and limited experience. Factors such as fatigue, varying skill levels, and subjective interpretation have led to certain rates of missed diagnoses and misdiagnoses, thereby limiting the effectiveness of early screening in cancer prevention and control. Furthermore, high-quality medical resources are primarily concentrated in first-tier cities, while remote areas suffer from a scarcity of medical resources.

Image source: iDeepWise

To address these challenges, iDeepWise has proposed a dedicated integrated solution combining AI algorithms with AI chips for the vertical scenario of medical imaging. The iDeepWise M-DPU is specifically optimized for medical imaging AI algorithms and boasts prominent advantages, including high computational power, compact size, low power consumption, rich expansion interfaces, ease of integration, and robust data privacy protection. It can be conveniently integrated into medical imaging equipment, enabling AI medical algorithm models to operate with high performance and high precision on medical terminal devices.Among these, the iDeepWise M-DPU provides an SDK for AI algorithms commonly used in medical imaging, significantly lowering the barrier to entry for smart healthcare AI and ultimately building a smart healthcare solution platform. Examples include common segmentation algorithms for overlapping cells and tissues, as well as prevalent medical imaging AI algorithms such as Faster R-CNN, SSD, FCN, and ResNet. The algorithmic component of iDeepWise’s current solutions encompasses two parts: medical image recognition in diagnosis and treatment, and natural language description and interaction with medical images. By leveraging deep learning artificial intelligence technology, it enables functions such as cancer screening (e.g., for cervical cancer) and medical consultations (including preoperative counseling and postoperative rehabilitation guidance).。

Image source: iDeepWise

Currently, the iDeepWise M-DPU has established collaborations with numerous large medical institutions and renowned Grade 3A hospitals, entering the stage of practical application. Extensive testing results demonstrate that this chip can intelligently classify 90,000 cells within 100 seconds. In ThinPrep Cytologic Test (TCT) applications, the iDeepWise M-DPU, combined with iDeepWise’s proprietary specialized algorithm for cervical cancer cell recognition, has delivered superior performance: the sensitivity for detecting squamous epithelial cell abnormalities reached 98.4%, with a specificity of 99.77%; the sensitivity for detecting glandular cell abnormalities reached 93.4%, with a specificity approaching 99%; and the accuracy of cell type classification reached 99.3%, surpassing the benchmarks set by the U.S. National Institutes of Health (NIH).

Image source: iDeepWise

In the future, iDeepWise will continue to deepen its engagement in the field of medical AI, bringing greater benefits to doctors and patients through pre-consultation doctor-patient interactions, intra-consultation medical imaging screening and recognition, and post-consultation follow-up and health consultations.