Abstract Background: A subset of EGFR-mutant non-small cell lung cancers (NSCLC) progresses to small cell lung carcinoma (SCLC) during tyrosine kinase inhibitor (TKI) therapy, particularly in cases with inactivating RB1 mutations. As RB1 mutations are likely foundational events in SCLC transformation, we hypothesize that tumors susceptible to this transformation exhibit morphologic features resembling SCLC even at the time of the initial diagnostic biopsy. Methods: We developed a predictive model for identifying the SCLC phenotype using H&E-stained slides from cases with confirmed SCLC histological diagnosis. Morphological features of individual cells were extracted through a grid-based image retrieval method, enhanced by subcellular compartmental analysis with an emphasis on nuclear and cytoplasmic characteristics. These features were used to quantify the morphological SCLC-likeness for each case. The SCLC-like phenotype was defined as the top 25% of cases exhibiting the highest SCLC-likeness scores. The model was then applied to 106 real-world advanced-stage EGFR-mutant NSCLC cases to validate the extracted features and assess clinical correlations. The primary endpoints were progression-free survival (PFS) after TKI therapy according to SCLC-like phenotype versus others, along with the rate of SCLC transformation. Secondary endpoints were RB1 mutations confirmed by targeted panel sequencing or whole exome sequencing. Results: The SCLC-like phenotype case images correlated with pathologist-interpretable features of SCLC, including a significantly smaller nuclear area (56 μm2 vs. 102 μm2) and increased nuclear optical density (18.01 vs. 15.68) reflecting hyperchromasia, consistent with typical SCLC morphology. Among the 106 patients with EGFR-mutant NSCLC, 26 were classified as having the SCLC-like phenotype. Patients with the SCLC-like phenotype demonstrated significantly worse PFS after TKI treatment (HR 1.71, CI 1.08-2.71, P = 0.02). Among the cases with secondary tissue biopsies (n = 68), SCLC transformation was observed in 4 cases, with a numerically higher frequency in the SCLC-like phenotype group compared to others (15.8% [3/19] vs. 2.0% [1/49], P = 0.06). No significant difference in RB1 mutation frequency was found via panel or whole exome sequencing. Conclusion: This is the first study to demonstrate the potential of morphologic assessment in identifying features pertinent to SCLC transformation and in optimizing prognostication in EGFR-mutant NSCLC treated with TKI. Given the heterogeneity in advanced-stage EGFR-mutant NSCLCs and the need to optimize therapeutic regimens, further and prospective studies are warranted. Citation Format: Soohyun Hwang, Hyukjung Kim, Yeong Hak Bang, Jun-Gi Jeong, Chang Ho Ahn, Seungeun Lee, Sanghoon Song, Aaron Valero Puche, JaeWoong Shin, Sehhoon Park, Hyun Ae Jung, Jong-Mu Sun, Yoon-La Choi, Jin Seok Ahn, Myung-Ju Ahn, Siraj Ali, Chan-Young Ock, Se-Hoon Lee. AI-powered assessment of morphologic likeness to small cell lung cancer (SCLC) predicts progression to SCLC and TKI response in EGFR-mutant NSCLC [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4659.
Read full abstract