You can deploy this model and start earning money today!

skblv/
resnet50-dsad-anatomy

Name: resnet50-dsad-anatomy
Availability: InStock

multi-label-classification

image-classification

apache-2.0

Views

No views yet

ResNet-50 — DSAD anatomical structure recognition

Supervised multi-label ResNet-50 baseline for recognizing which of 12 anatomical structures are visible in laparoscopic frames from the Dresden Surgical Anatomy Dataset (DSAD).

Trained as a baseline for the SDSC × Chicago Booth surgical video understanding leaderboard (Anatomy tab).

Prompt example

This closed-set example mirrors the leaderboard format, not a text-input API for this checkpoint.

text

[surgical frame]

Which anatomical structures are visible in this laparoscopic frame?
Select every matching label.

- abdominal wall
- colon
- inferior mesenteric artery
- intestinal veins
- liver
- pancreas
- small intestine
- spleen
- stomach
- ureter
- uterus
- vesicular glands

Model

torchvision.models.resnet50 initialized from IMAGENET1K_V2 weights
Final fc replaced with a 12-way linear layer, trained with BCEWithLogitsLoss (multi-label)
224×224 inputs, batch size 32, seed 42; full training code in s67_dsad_supervised.py

Evaluation

Full 1,978-frame video-level validation split (95% bootstrap CI):

Metric	Value
Exact match (predicted set == ground-truth set)	30.6% (28.5–32.6)
Micro-averaged F1	71.6% (70.5–72.8)

Per-class scores are in class_metrics.csv. This small supervised baseline beats all zero-shot vision-language models tested on this task; see the leaderboard.

Usage

python

import torch
from torchvision import models

model = models.resnet50()
model.fc = torch.nn.Linear(model.fc.in_features, 12)
model.load_state_dict(torch.load("best_model.pt", map_location="cpu"))
model.eval()
# sigmoid(logits) > 0.5 per class; class order in class_metrics.csv

References

Skobelev, K., Fithian, E., Baranovski, Y., et al. A Comparative Study in Surgical AI: Potential and Limitations of Data, Compute, and Scaling. arXiv:2603.27341, 2026.
Dataset: Carstens, M., Rinner, F. M., Bodenstedt, S., et al. The Dresden Surgical Anatomy Dataset for Abdominal Organ Segmentation in Surgical Data Science. Scientific Data, 10, 3 (2023).

Limitations

Research baseline only. Not a medical device. Trained on a single dataset (robot-assisted rectal resection); expect degraded performance on other procedures, optics, or institutions.

skblv/resnet50-dsad-anatomy