The knee joint, a complex hinge joint, is one of the primary load-bearing joints in the human body, facilitating various movements required for daily activities. Aging and injury can cause several knee abnormalities, leading to pain and dysfunction. Therefore, accurate diagnosis of these abnormalities is essential for creating tailored treatment plans and enhancing patients' quality of life. Given the knee joint's intricate anatomy, different scanning parameters can yield varied results. Additionally, some subtle lesions may be overlooked, especially by radiologists with limited experience. Multi-sequence knee magnetic resonance imaging (MRI) is an advanced, non-invasive technique for diagnosing knee pathologies. However, interpreting MRI results is time-consuming and heavily dependent on the expertise of the radiologist. In response to this, researchers have developed an innovative deep learning model that assists in classifying 12 common types of knee abnormalities, improving both efficiency and diagnostic accuracy.

In a collaborative study, researchers at HKUST Smart Lab (Kowloon, Hong Kong) and the Third Affiliated Hospital of Southern Medical University (Guangzhou, China) analyzed data from 1,748 patients. The study included T1-weighted (T1W), T2-weighted (T2W), and proton density-weighted (PDW) MRI sequences from sagittal, coronal, and axial planes. By combining these MRI data with arthroscopy results, which is considered the gold standard for diagnosing knee abnormalities, the researchers conducted a comprehensive analysis and identified 12 common knee abnormalities. They then developed a deep learning model, incorporating Co-Plane Attention across MRI Sequences (CoPAS), to classify these abnormalities. The model effectively captured intensity variations across different MRI sequences and identified complex correlations with abnormality types by separating spatial features, which resulted in high classification accuracy.

To assess the model's effectiveness, the researchers conducted simulated clinical testing. Radiologists first provided diagnoses based solely on MRI scans. After a washout period, the radiologists were asked to make diagnoses again, this time using the model’s output as a reference. The results showed that the model's average diagnostic accuracy exceeded that of junior radiologists and was comparable to senior radiologists. The overall diagnostic accuracy of all radiologists improved significantly with the model's assistance, according to the study published in Nature Communications. An additional interpretability analysis compared the clinical findings with the model’s output, revealing that the model’s decision-making process closely matched clinical preferences. This suggests that the model has developed a decision-making framework similar to that of radiologists, which enhances its reliability for clinical implementation.

“This innovative CoPAS model demonstrates diagnostic performance comparable to that of radiologists. It is particularly beneficial in bridging the gap between less experienced and senior doctors,” said Assistant Professor CHEN Hao from the Department of Computer Science and Engineering and Department of Chemical and Biological Engineering at HKUST, who led the study. “Our findings underscore the promise of artificial intelligence in healthcare, highlighting its potential to identify and validate new clinical insights.” 

http://m.allmycbs.com/en/index.asp .