Abstract:

Cutaneous T-cell lymphomas (CTCL) represent the majority of cutaneous lymphomas and comprise the second most common group of extranodal lymphomas. CTCL may show overlapping features with spongiotic dermatitis and distinguishing between the two may be diagnostically challenging. The histologic diagnosis is made using a combination of morphologic, immunophenotypic, and molecular features. We aim to determine whether immunophenotypic features of CTCL can be automatically quantified using artificial intelligence algorithms applied to digital images. Archival slides of spongiotic dermatitis and cutaneous T-cell lymphomas (N=70) stained for CD3, CD4, CD8 and reported by the University of Pittsburgh Department of Dermatopathology from 2012-2022 were scanned and converted to whole slide images using an Aperio AT2 DX System. Using QuPath 0.3, lymphocytes were automatically detected and scored based on expression of each of the immunohistochemistry stains. A CD4/CD8 ratio of >2 was used to determine if cases demonstrated an abnormal increase in CD4+ T-cells. A diagnosis of lymphoma vs benign was rendered on each case based on the patient’s history in combination with the presence or absence of T-cell clonality. CTCL cases consisted of CD4+ CTCL: Mycosis Fungoides (n=18), Anaplastic Large Cell Lymphoma (n=6), Small/Medium Peripheral T-cell Lymphoma (n=2), Sezary Syndrome (n=1), CD8+ CTCL: Subcutaneous Panniculitis-like T-cell Lymphoma (n=2), and CD4/8+: CTCL NOS (n=8). Spongiotic dermatitis (n=33) comprised the remainder of cases. Of the CD4+ CTCL, 19 (73%) of cases showed CD4/CD8>2. For spongiotic dermatitis, 24 (73%) of cases showed CD4/CD8>2. While both CD4+ CTCL and benign cases both showed elevated levels of CD4+ expression, our study shows that immunoquantification can be reliably automated using AI and digital pathology. AI tools can be used to potentially flag cases showing abnormal immunophenotypes before the pathologist signs out the case as an additional safe check.

Financial Disclosure:
No current or relevant financial relationships exist.

Christina Ragnarök

Affiliation

Refine Results

(Poster #380) Automated Immunophenotyping of Cutaneous T-cell Lymphoma Using Artificial Intelligence