Alvin Lang
September 6, 2024 13:46
A groundbreaking AI-based pathology platform developed by researchers at the University of Cologne provides accurate lung cancer diagnosis and treatment.
A recent study has introduced a cutting-edge AI-based pathology platform designed to help diagnose and evaluate lung cancer with unprecedented speed and accuracy. Developed by researchers at the University of Cologne Medical School and Cologne University Hospital, the new tool provides fully automated, in-depth analysis of benign and malignant tissue, paving the way for faster, more personalized treatment.
Innovation in Lung Cancer Diagnosis
Lung cancer, notorious for its high mortality rate, often benefits from accurate diagnosis and personalized treatment. Traditionally, oncologists manually examined tissue samples under a microscope to identify cancer cells. However, this process is time-consuming, subjective, and variable, sometimes leading to misdiagnosis.
To address these challenges, researchers developed a deep learning-based multi-class tissue segmentation platform that automatically analyzes digitized lung tissue samples, not only screening for cancer but also providing detailed cellular information about the examined area.
Advanced AI Training and Validation
The AI model was trained and validated on a substantial dataset of 4,097 annotated slides from 1,527 patients across six institutions. According to Yuri Tolkach, the study’s lead author, “The algorithm can distinguish 11 tissue types, ranging from tumor tissue and tumor-related classes to cartilage and lymphoid tissue. It showed very high pixel-by-pixel accuracy for segmenting the different classes, with an average Dice Score of 0.893.”
The researchers leveraged the University of Cologne’s high-performance computing cluster, which features 12 NVIDIA V100 GPUs, four NVIDIA A100 GPUs on an AI server in the Pathology Institute, and PC stations with NVIDIA GeForce RTX 3090 and RTX 4090 GPUs. This setup allows for rapid analysis of whole-slide images, which can take between one and five minutes per image and range in size from 200 to 2,000 MB.
Implications for cancer treatment
Beyond diagnosis, AI tools can reveal intricate details about tumors and immune cells within the cellular environment, providing insight into how cancer interacts within the body. Identifying subtle patterns and relationships within tissue samples that are invisible to the naked eye can lead to more accurate and effective treatments and a better understanding of how patients respond to specific cancer treatments.
“The composition of our research group and the results of our first large-scale cancer study Nature Machine Intelligence “This was made possible by a NVIDIA Quadro P6000 GPU grant from the NVIDIA Academic Grants Program,” Tolkach added.
The code used in this study is available on GitHub. The full study title is: Next-generation lung cancer pathology: development and validation of diagnostic and prognostic algorithmsYou can access it here.
Image source: Shutterstock