MLSP 2014 Schizophrenia Classification Challenge¶

For details, see https://www.kaggle.com/c/mlsp-2014-mri/overview

submission #842079¶

In [1]:

import numpy as np
import pandas as pd
from os import path

Data¶

For more details, see https://www.kaggle.com/c/mlsp-2014-mri/data.

About FNC Features¶

Functional Network Connectivity (FNC) are correlation values that summarize the overall connection between independent brain maps over time. Therefore, the FNC feature gives a picture of the connectivity pattern over time between independent networks (or brain maps). The provided FNC information was obtained from functional magnetic resonance imaging (fMRI) from a set of schizophrenic patients and healthy controls at rest, using group independent component analysis (GICA). The GICA decomposition of the fMRI data resulted in a set of brain maps, and corresponding timecourses. These timecourses indicated the activity level of the corresponding brain map at each point in time. The FNC feature are the correlations between these timecourses. In a way, FNC indicates a subject's overall level of 'synchronicity' between brain areas. Because this information is derived from functional MRI scans, FNCs are considered a functional modality feature (i.e., they describe patterns of the brain function). More about FNCs can be found here: FNC paper.

In [2]:

FNC_train = pd.read_csv(path.join('Train', 'train_FNC.csv'), index_col=0)

About SBM Loadings¶

Source-Based Morphometry (SBM) loadings correspond to the weights of brain maps obtained from the application of independent component analysis (ICA) on the gray-matter concentration maps of all subjects. Gray-matter corresponds to the outer-sheet of the brain; it is the brain region in which much of the brain signal processing actually occurs. In a way, the concentration of gray-matter is indicative of the "computational power" available in a certain region of the brain. Processing gray-matter concentration maps with ICA yields independent brain maps whose expression levels (i.e., loadings) vary across subjects. Simply put, a near-zero loading for a given ICA-derived brain map indicates that the brain regions outlined in that map are lowly present in the subject (i.e., the gray-matter concentration in those regions are very low in that subject). Because this information is derived from structural MRI scans, SBM loadings are considered a structural modality feature (i.e., they describe patterns of the brain structure). More about SBM loadings can be found here: SBM paper.

In [3]:

SBM_train = pd.read_csv(path.join('Train', 'train_SBM.csv'), index_col=0)

Labels for the training set.¶

The labels are indicated in the "Class" column. 0 = 'Healthy Control', 1 = 'Schizophrenic Patient'

In [4]:

labels = pd.read_csv(path.join('Train', 'train_labels.csv'), index_col=0)

Feature selection¶

In [5]:

%matplotlib inline
import matplotlib.pyplot as plt

In [6]:

from sklearn.feature_selection import SelectKBest, f_classif

In [7]:

FNC_selector = SelectKBest(f_classif).fit(FNC_train, labels['Class'])
SBM_selector = SelectKBest(f_classif).fit(SBM_train, labels['Class'])

In [8]:

fig = plt.figure(figsize=(18,6))
plot = fig.add_subplot(111, xlabel='feature number', ylabel='score', title='Univariate FNC feature scores')
indices = np.arange(FNC_train.shape[-1])
FNC_scores = -np.log10(FNC_selector.pvalues_)
FNC_scores /= FNC_scores.max()
plot.plot(indices, FNC_scores, 'r^')
plot.vlines(indices, [0], FNC_scores)
plt.show()

In [9]:

fig = plt.figure(figsize=(18,6))
plot = fig.add_subplot(111, xlabel='feature number', ylabel='score', title='Univariate SBM feature scores')
indices = np.arange(SBM_train.shape[-1])
SBM_scores = -np.log10(SBM_selector.pvalues_)
SBM_scores /= SBM_scores.max()
plot.plot(indices, SBM_scores, 'b^')
plot.vlines(indices, [0], SBM_scores)
plt.show()

For this submission I selected 60 FNC features and 8 SBM features.

In [10]:

FNC_selector.k = 60
SBM_selector.k = 8

In [11]:

FNC_features = FNC_train.columns[FNC_selector.get_support()]
SBM_features = SBM_train.columns[SBM_selector.get_support()]

In [12]:

X_train = pd.concat([ FNC_train[FNC_features], SBM_train[SBM_features]], axis=1)
y_train = labels['Class']

In [13]:

X_train

Out[13]:

	FNC13	FNC30	FNC33	FNC35	FNC37	FNC38	FNC40	FNC41	FNC42	FNC43	...	FNC353	FNC368	SBM_map7	SBM_map17	SBM_map36	SBM_map52	SBM_map61	SBM_map64	SBM_map67	SBM_map75
Id
120873	0.270490	0.036615	0.21516	0.069346	-0.086613	0.054857	-0.365520	-0.273410	-0.275500	-0.035595	...	-0.23049	-0.060204	-0.264192	0.137624	-1.062109	0.791762	-0.982331	1.070363	0.220316	-0.002006
135376	-0.088119	0.450290	0.70298	0.543640	0.244000	0.512400	0.439300	0.125780	0.191420	-0.058085	...	-0.31699	0.369300	-0.466051	0.972934	0.044317	-0.073326	-0.057543	0.371701	-0.513081	-0.295125
139149	-0.361020	0.203270	0.51565	0.114280	0.262910	0.018740	0.088855	-0.211420	0.026982	0.093953	...	-0.17144	0.521330	1.439242	-1.488153	0.414747	-0.910225	0.597229	1.220756	-0.059213	0.350434
146791	-0.060777	0.658060	0.64302	0.589520	0.485810	0.574150	0.344040	0.255670	0.091637	0.183470	...	0.27329	0.144460	-0.492673	0.187573	-0.026555	-3.013096	0.829697	-0.450726	-0.791032	0.448966
153870	0.048705	0.158000	0.25707	0.152580	-0.105510	-0.234190	-0.127320	0.143880	-0.286530	-0.333980	...	-0.49929	-0.179310	-1.105922	1.961955	-1.027496	0.474353	-0.978412	0.158492	0.889753	-0.551440
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
934330	-0.373800	0.706610	0.53835	0.428510	0.533970	0.317520	-0.042524	0.249370	0.191810	0.184660	...	-0.46915	0.108670	-0.515736	-0.112326	0.458916	-0.516803	0.826910	-0.225792	0.369724	0.567717
950671	-0.029000	0.516310	0.41210	0.354370	0.311470	-0.067289	0.011223	0.173140	-0.039187	0.137150	...	-0.42424	0.304690	-0.933527	-0.347191	0.183240	0.914623	0.482055	0.073327	-0.455141	0.977018
963924	-0.197060	0.011246	0.35735	0.623080	0.317980	-0.081810	-0.102280	0.029382	0.205780	0.374430	...	0.42983	0.357230	-0.523021	1.369376	-0.976704	-1.429466	-0.114078	-0.476524	-0.556896	-0.424864
993348	-0.087478	0.420390	0.29361	0.026402	-0.041024	0.355390	0.163750	-0.186250	-0.297480	0.179980	...	0.15890	0.462660	0.462689	-1.749746	-0.385862	0.839745	-0.163926	0.953385	0.402673	-0.421040
993946	-0.061554	-0.125080	0.15613	0.269830	-0.178200	-0.133630	0.019102	0.383180	-0.082001	-0.058647	...	-0.17677	0.307260	1.522190	-0.281454	-0.907096	1.241959	-0.626201	0.229092	1.358773	-1.125141

86 rows × 68 columns

In [14]:

labels

Out[14]:

	Class
Id
120873	1
135376	0
139149	0
146791	0
153870	1
...	...
934330	0
950671	0
963924	1
993348	0
993946	1

86 rows × 1 columns

Training¶

In [15]:

from sklearn.neighbors import KNeighborsClassifier

In [16]:

clf = KNeighborsClassifier(n_neighbors=35).fit(X_train, y_train)

In [17]:

print(clf.score(X_train, labels))

0.872093023255814

Submission¶

In [18]:

FNC_test = pd.read_csv(path.join('Test', 'test_FNC.csv'), index_col=0)

In [19]:

SBM_test = pd.read_csv(path.join('Test', 'test_SBM.csv'), index_col=0)

In [20]:

X_test = pd.concat([FNC_test[FNC_features], SBM_test[SBM_features]], axis=1)
X_test

Out[20]:

	FNC13	FNC30	FNC33	FNC35	FNC37	FNC38	FNC40	FNC41	FNC42	FNC43	...	FNC353	FNC368	SBM_map7	SBM_map17	SBM_map36	SBM_map52	SBM_map61	SBM_map64	SBM_map67	SBM_map75
Id
100004	0.113166	0.304551	0.678723	0.676364	0.102174	0.022148	0.317406	0.464976	0.236121	-0.047108	...	-0.483997	0.150698	-2.404130	2.256762	-2.011786	0.139369	1.123770	2.083006	1.145440	0.192076
100015	-0.054457	0.315034	0.770686	0.787717	0.366940	-0.299074	0.699394	0.452051	0.032888	0.262658	...	-0.252089	0.318086	-0.612468	1.711094	0.185261	-2.084801	1.397832	1.046136	-0.191733	0.174160
100026	0.002372	-0.108333	-0.058818	0.316538	0.081580	0.113453	-0.050571	0.068699	0.275885	0.562116	...	0.294130	0.093333	-0.752907	1.386814	-0.123830	0.046525	1.906989	-2.661633	-0.193911	-0.476647
100030	0.040945	0.675230	0.537128	0.124338	0.073727	0.278834	0.161171	0.316369	0.067719	-0.046476	...	-0.270263	0.442313	1.041755	-0.949757	0.167515	-1.693663	-1.997087	-2.083782	1.154107	2.790871
100047	-0.245284	0.624100	0.294196	0.332166	0.447499	0.264495	0.446376	0.400933	0.041004	0.538643	...	-0.204068	0.394492	-1.775324	0.415556	2.410666	0.021838	1.578984	1.402592	-1.230440	-1.544345
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
999956	0.182524	0.241864	0.606688	-0.000529	0.566684	-0.232798	0.733727	-0.056436	-0.227445	0.171759	...	-0.184820	0.407409	-1.123315	0.428509	2.027816	0.849052	1.079632	-1.525021	2.479249	0.548342
999979	-0.172684	0.259993	0.571281	-0.075842	-0.262201	0.019818	0.504592	-0.185592	-0.809122	0.389232	...	-0.310955	0.473550	-1.270540	0.013607	-0.047793	-0.755896	0.192909	-0.336587	0.062809	-0.690686
999989	-0.408067	0.862003	0.646698	0.700784	0.572094	-0.678907	0.192104	-0.556889	-0.246365	-0.199630	...	-0.167875	-0.437320	-0.501659	-0.680883	1.595677	0.466304	0.430025	1.452216	0.759654	1.607317
999992	-0.160676	-0.055138	0.247362	0.478874	0.195440	0.358874	-0.124268	0.284184	0.330500	-0.258015	...	-0.497435	0.783766	-1.248628	0.815399	1.427952	-0.305261	-2.599313	-0.161822	0.183669	-1.313079
999994	0.005356	0.403151	0.538632	0.792229	-0.226379	-0.229716	-0.044973	-0.262501	-0.039308	-0.207124	...	0.300068	0.562554	-1.980608	-1.398193	-1.263830	-0.303426	0.785046	2.369759	-0.168919	-0.668412

119748 rows × 68 columns

In [21]:

y_test = clf.predict_proba(X_test)
submission = pd.DataFrame(y_test, index=X_test.index, columns=['Healthy Control', 'Schizophrenic Patient'])
submission

Out[21]:

	Healthy Control	Schizophrenic Patient
Id
100004	0.342857	0.657143
100015	0.542857	0.457143
100026	0.600000	0.400000
100030	0.685714	0.314286
100047	0.600000	0.400000
...	...	...
999956	0.485714	0.514286
999979	0.428571	0.571429
999989	0.600000	0.400000
999992	0.342857	0.657143
999994	0.514286	0.485714

119748 rows × 2 columns

In [22]:

submission.to_csv('submission.csv', columns=['Schizophrenic Patient'], header=['Probability'])