14-MONTH ML RESEARCH INTERNSHIP · ANRF-FUNDED

Oral cancer
histopathology
classification

Deep learning for oral-cancer histopathology and WPOI classification.

During this 14-month ANRF-funded internship, I worked with 292 whole-slide images, developed a ResNet-18-based model in PyTorch, and built a Flask demo for inference.

MICROSCOPE FIELD / 01CONCEPTUAL RENDER · NO PATIENT DATA
ABSTRACT FIELDFOCUS / 01
F_07
F_12
INPUT / H&E IMAGE
IMAGE → PREP → MODEL → RESULTFOCUS PLANE / 01
01 / IMAGEOral-cancer histopathology · WPOI classification.

CONCEPTUAL VISUAL · NO PATIENT DATA OR PROJECT SLIDE SHOWN

THE RESEARCH WORKFLOW

From image
to demonstration.

  1. 01

    292 whole-slide images

    Oral-cancer histopathology · WPOI classification.

  2. 02

    Preprocess & tile

    Whole-slide images prepared as model inputs.

  3. 03

    Train the model

    ResNet-18-based deep learning with PyTorch.

  4. 04

    Evaluate

    ~91.47% accuracy · macro-F1 ~0.91 · ~21-point lift over baseline.

  5. 05

    Demonstrate

    Prediction through a Flask inference application.

RESEARCH RECORD

What came out of the work.

292WHOLE-SLIDE IMAGES
~91.47%PROJECT EVALUATION ACCURACY
~0.91MACRO-F1
~21 ppIMPROVEMENT OVER BASELINE
Conceptual microscope slide artwork in rose and violet tones
01 / IMAGE INPUTCONCEPTUAL ARTWORK
Conceptual research desk showing a deep-learning image workflow
02 / MODEL WORKFLOWCONCEPTUAL ARTWORK

Reported project results; they are not independent clinical validation results.

Technical details

Model: ResNet-18-based deep-learning model with custom feature modulation and angular-margin loss.

Workflow: Whole-slide preprocessing and tiling → model training → evaluation → Flask inference.

PUBLICATION

One Elsevier publication related to this research.

PATENT APPLICATION

An application associated with the project.

The Springer Nature 2025 chapter Computational Methods in Oncology is related work and is listed separately.