Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-task learning for classification, segmentation, reconstruction, and detection on chest CT scans

Weronika Hryniewska-Guzik, Maria Kędzierska, Przemysław Biecek

TL;DR

The paper addresses the challenge of identifying lesions on chest CT scans for lung cancer and COVID-19 by proposing a four-task multi-task framework that jointly handles classification, segmentation, reconstruction, and detection. It builds a U-net–based model with a shared encoder and four task-specific decoders, including a Mask-RCNN detection head, and trains with a joint loss $\mathcal{L}_{total}= w_1\cdot \mathcal{L}_{classif}+ w_2\cdot \mathcal{L}_{segm}+ w_3\cdot \mathcal{L}_{recon}+ w_4\cdot \mathcal{L}_{detect}$ where $w_i \in \{0,1\}$; segmentation uses generalized Dice loss. The study uses 2D CT slices from public datasets, tests two backbones, and reports that auxiliary tasks can improve classification and segmentation while detection remains data-limited, highlighting the need for more annotated data and augmentation. Overall, the work demonstrates feasibility of adding detection to multi-task learning for chest CT and points to practical benefits and remaining data-driven challenges for robust lesion detection.

Abstract

Lung cancer and covid-19 have one of the highest morbidity and mortality rates in the world. For physicians, the identification of lesions is difficult in the early stages of the disease and time-consuming. Therefore, multi-task learning is an approach to extracting important features, such as lesions, from small amounts of medical data because it learns to generalize better. We propose a novel multi-task framework for classification, segmentation, reconstruction, and detection. To the best of our knowledge, we are the first ones who added detection to the multi-task solution. Additionally, we checked the possibility of using two different backbones and different loss functions in the segmentation task.

Multi-task learning for classification, segmentation, reconstruction, and detection on chest CT scans

TL;DR

The paper addresses the challenge of identifying lesions on chest CT scans for lung cancer and COVID-19 by proposing a four-task multi-task framework that jointly handles classification, segmentation, reconstruction, and detection. It builds a U-net–based model with a shared encoder and four task-specific decoders, including a Mask-RCNN detection head, and trains with a joint loss where ; segmentation uses generalized Dice loss. The study uses 2D CT slices from public datasets, tests two backbones, and reports that auxiliary tasks can improve classification and segmentation while detection remains data-limited, highlighting the need for more annotated data and augmentation. Overall, the work demonstrates feasibility of adding detection to multi-task learning for chest CT and points to practical benefits and remaining data-driven challenges for robust lesion detection.

Abstract

Lung cancer and covid-19 have one of the highest morbidity and mortality rates in the world. For physicians, the identification of lesions is difficult in the early stages of the disease and time-consuming. Therefore, multi-task learning is an approach to extracting important features, such as lesions, from small amounts of medical data because it learns to generalize better. We propose a novel multi-task framework for classification, segmentation, reconstruction, and detection. To the best of our knowledge, we are the first ones who added detection to the multi-task solution. Additionally, we checked the possibility of using two different backbones and different loss functions in the segmentation task.
Paper Structure (7 sections, 2 equations, 4 figures, 3 tables)

This paper contains 7 sections, 2 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Multi-task model training on CT chest scans
  3. Data and preprocessing
  4. Multi-task architecture
  5. Model training and results
  6. Evaluation on different backbone
  7. Conclusions

Figures (4)

  • Figure 1: Multi-task architecture and training diagram for tasks: classification (C), segmentation (S), reconstruction (R), and detection (D).
  • Figure 2: Masks generated by concurrent segmentation & reconstruction task with and without loading weights from classification task.
  • Figure 3: Evaluation of the loss during training detection & reconstruction task as a function of epoch.
  • Figure 4: Using two different backbones: VGG/̄13 and ResNet/̄50 for training classification & reconstruction (CR) and segmentation & reconstruction (SR) models.