Early diagnosis and treatment of cancer are critical for improving patient outcomes. However, traditional interpretation of immunohistochemical stained tissue slides, which relies on pathologists using microscopes, is time-consuming and subject to human error due to subjective factors. Professor Chia Yen Lee’s team from the Department of Electrical Engineering at National United University (NUU) has developed a solution to this challenge with the creation of the "Ki-67 Automatic Interpretation Model for Neuroendocrine Tumor Immunohistochemical Stained Slides" (KissNET). This innovation marks a revolutionary shift in pathological diagnosis.
Combining Deep Learning with Cutting-Edge Technology: KissNET Delivers Accurate Interpretations
Ki-67 stained slides are used to assess tumor proliferation rates and have become a key indicator for cancer prognosis. However, variability in tissue morphology, uneven staining, and diverse cell types make manual interpretation slow and prone to challenges, a global issue faced in pathology today. KissNET, driven by deep learning technology, addresses this by training on a large dataset of pathological images, allowing the model to automatically recognize the features of tumor cells. KissNET integrates two major innovative technologies: tumor region segmentation and cell counting.
Tumor Region Segmentation: Traditional interpretation relies on the naked eye to identify tumor regions, which can be affected by tumor heterogeneity. KissNET enhances model depth while retaining image detail, enabling more precise automatic segmentation of tumor regions, excluding interference, and improving interpretation accuracy.
Cell Counting: The varied and often overlapping cell types seen in Ki-67 stained slides make manual cell counting difficult. KissNET employs an attention mechanism, improves loss function, and strengthens the segmentation network, effectively identifying and counting cells, overcoming the limitations of traditional methods.
Professor Chia Yen Lee: KissNET Speeds Up Diagnosis and Enhances Healthcare Quality
Professor Chia Yen Lee highlighted that the time required for interpretation using KissNET has been significantly reduced from 8.1 minutes with manual interpretation to just 2 seconds—243 times faster. This not only greatly shortens the time needed for interpretation but also enhances objectivity and consistency, providing pathologists with substantial support, allowing them to focus on complex cases and research. She emphasized that KissNET has already been deployed in clinical settings at Taipei Veterans General Hospital and Taichung Veterans General Hospital, where its predictions align closely with pathologists' interpretations, demonstrating its clinical value.
Looking forward, the team plans to customize AI models based on different clinical environments, integrating both hardware and software systems to maintain the platform's stable operation. The goal is to extend the system's applicability to other types of cancer diagnoses. Additionally, KissNET’s user-friendly visual interface will be further refined to help physicians better understand the results, enhancing their trust in AI technology.
Professor Lee stressed that KissNET’s mission is not just to speed up the diagnostic process but also to improve the accuracy and efficiency of pathological diagnoses through AI, ultimately providing patients with higher-quality medical services. KissNET is set to become an indispensable tool for pathologists, driving breakthroughs in cancer diagnosis and treatment.
Resource (mandarin): AI辨識腫瘤細胞特徵 精準判讀癌症預後指標
