AI-based morphological characterization of small cell lung cancer subtyped by molecular transcriptional pattern bei Technische Universität München

Background:
Small cell neuroendocrine carcinoma (NEC) is one of the most aggressive neuroendocrine neoplasms. Small cell lung cancer (SCLC) accounts for approx. 85% of small cell NEC and have a distinct clinical, morphomolecular feature from those with neuroendocrine tumors (NET). Previous molecular analyses have revealed several tumor groups within SCLCs. Recent studies identified four major transcriptional patterns of SCLCs, that differ in protein and RNA- expression patterns and drug-sensitivity. However, diagnostic significance of the transcriptionally defined groups is not known.

Aim: This study is designed to understand detailed morphology of SCLCs and to identify morphological and immunohistochemical features that correlate with newly classified transcriptional subtypes.

Patients´ characteristics: Surgically resected SCLC tissues obtained from 76 patients (median 71-year-old, 93% heavy smoker) were reviewed. Histologically, 61 (80%) were pure SCLCs, while 15 (20%) were combined SCLCs either with a component of squamous cell carcinoma or adenocarcinoma. The tumors were subtyped into three transcriptional groups; Type A (53%), Type N (25%), and Type P (20%), based on previously established immunohistochemical expression of ASCL1 for Type A, NeuroD1 for Type N and POU2F3 for Type P. Three percent of the tumors were negative for the three transcriptional factors. The median follow-up after surgery was 16.5 months. The overall survival (OS) and recurrence-free survival (RFS) rates were significantly associated with the TNM-stage (OS, p = 0.0333; RFS, p = 0.0063). No significant differences were seen in transcriptional subgroups (OS, p = 0.7804; RFS, p = 0.1131).

Methods and study plan: In order to assess precise morphology of the tumor tissues (e.g. growth pattern, cell density, cell size, nuclear shape, chromatin distribution pattern, number of mitosis, extent of necrosis, etc.) the HE-stained whole tissue slides were scanned by Nano Zoomer XR (Hamamatsu Photonics K.K., Hamamatsu, Japan). The above mentioned histological and cytological features of SCLC tissue will be analysed using digital tools such as QuPath (https://doi.org/10.1038/s41598-017-17204-5), TMARKER (https://doi.org/10.4103/2153-3539.109804) and deep learning based tools to extract pathomics (https://doi.org/10.1038/s41467-023-36173-0). Software skills should be in a script language (python or R). To clarify morphological characteristics in the transcriptional subtypes, morphometric data will be compared with the subtypes, distribution of the transcriptional protein expression as well as other immunohistochemical marker expression (i.e. Synaptophysin, chromogranin, CD56, INSM1, TTF1, SSTR2, Ki67).



Comments:
This is potentially an ideal study project for a master thesis for a student from the field of IT-science or AI-based medicine. A student with basic or advanced skills in programming is, therefore, desired. The computational pathology will be supervised by Prof. Peter Schüffler and a medical aspects and histomorphological pathology of the tumors will be supervised by Dr. med. Atsuko Kasajima and Prof. Günter Klöppel.

Kontakt: Dr. med. Ayako Ura, ayako.ito@tum.de