Lunit is highlighting new findings from colorectal cancer treatment research using the Lunit SCOPE HER2 analyzer, the SCOPE IO tumor microenvironment (TME) analyzer, and artificial intelligence (AI).
In a post hoc exploratory analysis of Triumph phase II trial outcomes, the AI analyzers significantly improved HER2 biomarker evaluation and the prediction of clinical outcomes in metastatic colorectal cancer (mCRC) in patients undergoing therapy targeting human epidermal growth factor receptor 2 (HER2), according to Lunit and researchers from Japan's National Cancer Center Hospital East.
The study, published on January 17 in JCO Precision Oncology, investigated the potential association between factors quantified by AI-powered whole-slide image (WSI) analyzers and clinical outcomes of dual therapy (trastuzumab and pertuzumab) in 30 patients with HER2-amplified mCRC.
(D and E) Representative patients from the Triumph study demonstrating AI-HER2 3+ with low (C, 18.1%) and high (D, 97.4%) AI-H3. The left panels show the original WSIs, and the right panels show the WSIs processed by the AI analyzer. Tumor cells are color-coded according to HER2 staining intensity. (Blue: HER2-negative tumor cells, green: HER2 1+ tumor cells, yellow: HER2 2+ tumor cells, red: HER2 3+ TC). The green-shaded area indicates the cancer area. Images courtesy of Imai et al. Licensed under CC BY 4.0.American Society of Clinical Oncology (ASCO)
Among the findings, SCOPE HER2 demonstrated 86.7% accuracy compared to pathologist assessments for HER2 immunohistochemistry (IHC), achieving 100% accuracy in identifying HER2 IHC 3+ cases, noted lead author Dr. Mitsuho Imai, PhD, from the National Cancer Center Hospital East in Chiba, Japan, and colleagues.
The researchers described the SCOPE HER2 analyzer as comprising two deep learning-based models, a cell detection model that categorizes five cell classes: H0, H1, H2, and H3 tumor cells, and other cells, and a tissue segmentation model.
Patients identified by the AI model as having a high proportion of HER2 IHC 3+ staining tumor cells (AI-H3-high, >50%) exhibited better clinical outcomes than those identified through traditional HER2 evaluation methods, Lunit said.
Outcomes results were measured in relation to an objective response rate (ORR) at 42.1% (AI-H3-high) versus 26.7% (overall Triumph trial), progression-free survival (PFS) at 4.4 months (AI-H3-high) versus 1.4 months (AI-H3-low), and overall survival (OS) at 16.5 months (AI-H3-high) versus 4.1 months (AI-H3-low).
The SCOPE IO analyzer runs on two deep learning-based AI models. Detailed TME profiling, including lymphocyte, macrophage, and fibroblast densities, found that AI-H3-high patients with low stromal TME density achieved the most favorable outcomes, reported as ORR: 57.1%, PFS: 5.6 months, and OS: 26.0 months, according to Imai and colleagues.
"This study highlights the potential of AI-powered [quantification continuous score] QCS and TME analysis in predicting treatment responses in patients with mCRC undergoing [trastuzumab plus pertuzumab] TP therapy," the authors wrote, adding that large-scale prospective studies are crucial to validate and expand the findings. "Despite promising results, the exploratory nature and study limitations necessitate further research."
Principal investigator Dr. Takayuki Yoshino, PhD, said the ability to more precisely stratify patients will lead to more personalized treatment options, improving outcomes for patients with HER2-positive metastatic colorectal cancer.